The LeanScale Podcast · Episode 106

How She Runs All of RevOps From the Terminal

Sarah Madden (Smadds) on the three buckets every skill falls into, the rule of three, and the infrastructure discipline underneath it

Sarah Madden · Head of Revenue Operations & Strategy · FERMÀT Hosted by Anthony Enrico
Published Updated 01:13:40 79 min read 15843 words
Revenue Operations AI in GTM GTM Strategy Sales Enablement
▸ Listen on Spotify Watch on YouTube Apple Podcasts
SummaryTakeawaysFrameworksQuotesAdviceAIEntitiesMetricsFAQTimelineTranscript
Executive Summary

The one-paragraph brief, extended

Why this conversation matters — and who should spend the hour.

Sarah Madden — Smadds across RevOps — started as employee #64 at Braze, cold-calling as a BDR. Then a waterfall chart in an all-hands rearranged her career. She can still describe the CRO walking through each floating bar, stopping at expansion and churn, and her realising the chart existed because she had put the right closed-lost reason on an opportunity. Warned that operations was a lot of Excel time, she thought that sounded better, not worse. Today she is Head of Revenue Operations & Strategy at FERMÀT and runs almost her entire job out of the terminal.

The origin of that is a budget conversation. As FERMÀT's first RevOps hire facing renewals with no tracking, she needed 500 contract PDFs — amendments, MSAs, renewals, every one named differently — backfilled into Salesforce, and asked for roughly $20,000 of contractor support. Her co-founder gave her a $200 Cursor licence instead. What made it work was that he then refused to help: prepping questions for their Friday session, she realised he would just tell her to ask the chat bot, started asking it herself, and cancelled the call. No assumptions, no crutch, so she asked every question honestly.

The organising system is three buckets and three rules. Skills either surface context, own an end-to-end process, or maintain infrastructure — that third one being her favourite precisely because she is not a software engineer, so she asked Claude what it needed from her and built skills to cover blind spots she could not name. The rules all centre on three: more than three tools means rethink the workflow; more than one person means it has breadth; say or do something more than three times and it should be a skill. And the honest listen is for the anxiety, not the task — nobody says they cannot get follow-ups out on time, they say they are up late on email and always feel behind.

The demonstrations are specific. A tool-evaluation skill now run by the head of marketing, which scrapes Notion, Slack and Ramp spend and contains a persona of Sarah that interrogates the budget owner — the first time she has shifted work toward stakeholders rather than away from them. A data-enrichment bake-off in twenty minutes instead of three days, run against vendor API keys from the terminal. A lead-list import that went from a full day to forty minutes, where the enriched free-email addresses she used to delete are the real prize. And a Pipeline Ops plugin where Claude surfaces evidence for each field rather than filling it, written so it cannot be copy-pasted.

Underneath sits infrastructure discipline: a status bar showing context usage because quality degrades past 50%, a wrap-and-pickup pair used around 250 times a month, and a core-tools tracker built after her chief-of-staff agent reported no new emails for four hours while Gmail was silently down. On team design, her view is that people follow pain, RevOps is really a data-integrity team good at automations, and the technical half may drift toward data while strategy and forecast management stay. She wants tools with an MCP or CLI — without one she will not even take the demo. Her closing argument: treat agentic infrastructure like a cornerstone system, with maintenance days and code freezes, the way you already treat Salesforce.

Key Takeaways

15 things worth stealing

The load-bearing ideas, each with the business implication and who should care.

01

A $200 Cursor licence replaced a $20,000 contractor budget — and being refused help was the unlock

Facing 500 inconsistently named contract PDFs to backfill into Salesforce, Sarah asked for contractor support and got a Cursor licence instead; she did not know what Cursor or terminal were. Her co-founder then refused to help — prepping questions for a Friday session, she realised he would just tell her to ask the chat bot, so she asked it herself and cancelled the call.

Why it matters: The constraint was never the work, it was the assumption about who could do it. Having no crutch meant she went in with no assumptions about how anything worked and asked every question honestly.

RevOps LeadersFoundersRevenue Executives
02

Skills fall into three buckets — and infrastructure is the one non-engineers skip

Surfacing context (things you could do yourself, more slowly), process (an end-to-end workflow the skill now owns), and infrastructure (a healthy terminal, folder structure and file hygiene). Sarah fed roughly 70 skills into Claude and asked it to categorise them. For the third bucket — she had one folder titled after her first project holding twelve months of everything — she asked Claude what it needed from her to work as expected and built skills from the answers.

Why it matters: Operators work better with tags than blank space, and "I don't know what I don't know" is solvable if you ask the model to specify its own requirements rather than guessing at best practice.

RevOps LeadersRevenue ExecutivesFounders
03

The rule of three: say or do it three times and it should be a skill

Alongside two companions — if a process spans more than three tools, rethink the workflow; if more than one person follows it, it has breadth worth capturing. Sarah says thinking this way for ten days will shock you with how much repeats.

Why it matters: Almost every part of a job could become a skill. She estimates she has built about 10% of hers — the ones she is most comfortable with and has most control over.

RevOps LeadersRevenue ExecutivesSales Leaders
04

Listen for the anxiety, not the task

Nobody says they cannot get their follow-ups out on time. They say they are up late working on email and always feel behind. That second sentence is the pain point worth taking to Claude.

Why it matters: The stated request and the real problem are different artefacts. Sarah's version is someone going into fourteen tools for a QBR deck — the anxiety in how they describe it is the signal.

RevOps LeadersRevenue ExecutivesSales Leaders
05

Slack is a data lake, not a messaging app

Once you can scrape it, Slack holds qualitative data that was previously inaccessible. Anthony's example is combining shared client Slack channels, a transcript warehouse and the project system into one customer-health workflow that classifies accounts by health.

Why it matters: That would previously have needed a data science team of twenty and a team of engineers. The qualitative caveat that used to be a 10% vibes section in a health score is now addressable.

RevOps LeadersCustomer SuccessRevenue Executives
06

The tool-eval skill shifts work to stakeholders rather than away from them

The head of marketing now kicks off a renewal evaluation herself. It scrapes Notion, Slack and Ramp spend, finds alternatives, and contains a persona of Sarah that interrogates the budget owner — are you sure you need this, is there another way — before producing a one-pager posted to the RevOps channel.

Why it matters: Sarah grew up in RevOps believing the job was making things maximally easy for sellers. This is the first time she has moved work toward stakeholders, and she says she has not yet decided how she feels about it.

RevOps LeadersRevenue ExecutivesFounders
07

A vendor bake-off in twenty minutes instead of days

Rather than hand-weighting a test spreadsheet, Sarah described the ICP mix, dirty-data proportion and international split to Claude and got the file. Then she used each vendor's API key from the terminal to run the comparison directly, cross-referenced Ramp for current spend, Notion for dependent documentation, and Slack for AEs complaining about data quality.

Why it matters: The vendor-run version takes three days and ends in a PDF walkthrough. The point is not speed alone — it is enriching the decision with evidence you would never otherwise assemble.

RevOps LeadersRevenue Executives
08

The free email addresses you delete may contain enterprise deals

Lead-list imports went from a full interrupted day to about forty minutes, with a RevOps analyst able to run four a day. But the bigger win is enriching the free-email registrations that used to get deleted because they could not be matched.

Why it matters: Even if AI took the same amount of time, you would still be leaving money on the table and missing depth of analysis. Speed is the least interesting part of the change.

RevOps LeadersMarketing LeadersRevenue Executives
09

Pipeline Ops surfaces evidence rather than filling fields

For economic buyer, champion or legal process, Claude returns four bullet points of times someone explicitly said it — written so it cannot be copy-pasted, with a rule preventing it. The human has to rewrite it.

Why it matters: Sarah is explicit that plenty of teams autofill successfully and FERMÀT chose not to. They are not afraid of slowing an agentic process down to keep human judgement in the loop.

RevOps LeadersSales LeadersRevenue Executives
10

If removing AI gives the same output, you only made it faster

Sarah's test for whether a workflow was genuinely redesigned. Redesign starts from the problem statement — a clean pipeline with minimal work on someone else and maximum accuracy — and produces capabilities that did not exist before, such as an end-of-day skill checking calendars for calls with no matching opportunity.

Why it matters: Anthony's echo: focus on effectiveness over efficiency. Many things do not need doing faster, but there is enormous room to do them better.

RevOps LeadersRevenue ExecutivesFounders
11

Never run past 50% context — and build wrap and pickup so clearing it costs nothing

A go-to-market engineer told Sarah quality degrades beyond 50%, so she added a status bar showing usage, session spend and the active model. The wrap skill summarises what was decided and built, updates a cache and writes a compact handoff; pickup resumes from it. She copies the handoff, clears context, pastes, and continues — using the pair around 250 times a month.

Why it matters: The visual cue exists because she knows she will otherwise forget. It also revealed which operations are expensive: scanning Notion spikes cost, writing to it does not.

RevOps LeadersRevenue Executives
12

Build a tracker for silent tool failures

Sarah's chief-of-staff agent reported no new emails for four hours; the Gmail connection had broken. She built a core-tools tracker so she knows when email, calendar, Notion, Slack or Granola go offline, plus a recurring connection heartbeat.

Why it matters: An agent reporting nothing and an agent unable to see anything are indistinguishable from the outside. Anthony immediately said he was stealing this one.

RevOps LeadersRevenue Executives
13

No MCP or CLI, no relationship

Sarah will not seriously consider a tool without one, regardless of category — she will not even take the demo. If she has to go into your tool to use it, she will not use it. She applies the same rule personally, having installed MCPs for her fitness tracker and music.

Why it matters: She also questions whether conversation intelligence needs video at all: the transcript is what her team actually uses, and she is not going to rebuild that category to compete.

RevOps LeadersRevenue ExecutivesFounders
14

RevOps is a data-integrity team that is good at automations

Sarah's framing to her own team. Everything the business runs on comes from a system, and those systems rest on the processes and rules RevOps sets. She expects a bifurcation: the technical half drifting toward data teams, the strategy, process, enablement and forecast-management half staying.

Why it matters: Go-to-market data is the messiest data there is — Anthony's point that having done it on hard mode, taking on cleaner ERP or product data is not a stretch. She expects enablement to have the next resurgence after RevOps.

RevOps LeadersRevenue ExecutivesFounders
15

Treat agentic infrastructure like Salesforce

Maintenance days, code-freeze periods, blackout dates — all of it will apply to company-wide agentic infrastructure and to each person's own operating system, because this is now a cornerstone of the tech stack.

Why it matters: It follows from building out loud in a flat organisation: nobody is checking on you, so you have to keep saying what you are spotting and building even when unsure anyone is listening.

RevOps LeadersRevenue ExecutivesFounders
Frameworks Discussed

6 named models

Every framework Jimmy names, defined and time-stamped.

The Three Skill Buckets

17:26

Surfacing context (information you could gather yourself, more slowly), process (an end-to-end workflow the skill now owns entirely), and infrastructure (maintaining the terminal, folder structure and the system itself).

Derived by feeding roughly 70 existing skills into Claude and asking it to categorise them. Sarah's point is that operators work better with tags than blank space, and each bucket describes a different kind of leverage.

The Rule of Three

19:09

If a process requires more than three tools, rethink the workflow. If more than one person follows it, it has breadth worth capturing. If you say or do something more than three times, it should be a skill.

Sarah says the third has lived rent-free in her head for two months, and that ten days of thinking this way will shock you. Almost every part of a job qualifies; she estimates she has built about 10% of hers.

Listen for the Anxiety

19:31

The stated problem and the real one differ. Nobody says they cannot get follow-ups out on time; they say they are up late on email and always feel behind. The second is the pain point to build against.

Sarah surfaces this in conversations with account managers and delivery staff — you can hear the anxiety in how a process is described, and that is what gets taken to Claude.

Evidence, Not Autofill

36:43

For each CRM field, the agent returns bullet points of times someone explicitly said the thing, written so it cannot be copy-pasted, with a rule preventing it — the human rewrites the entry.

A deliberate choice rather than a limitation. FERMÀT is not ready to autofill and is willing to slow the process to keep human judgement in the loop, while keeping a log of every field updated and why.

The Remove-the-AI Test

38:03

If you removed AI from a workflow and would still get the same output, AI only made it faster — you have not redesigned anything.

Redesign starts from a problem statement and produces capability that did not previously exist, such as an end-of-day scan of calendars for calls with no matching opportunity.

Wrap and Pickup

1:01:14

A wrap skill that summarises decisions, updates a cache and writes a compact handoff file at the end of a session, paired with a pickup skill that resumes exactly where it left off after clearing context.

Built to keep every session on the strongest version of the model rather than degrading past 50% context. Used about 250 times a month across work and personal projects.

Best Quotes

39 lines worth clipping

Pulled verbatim. Copy or share any of them.

“You can hear it. You can hear the anxiety. That's the problem. That's the pain point that we're solving.”
Sarah Madden 00:28
“People say, "I'm up so late at night working on my email, and I feel like I'm always behind." That's the pain point.”
Sarah Madden 00:36
“If you say something or you do something more than three times, it should be a skill.”
Sarah Madden 01:07
“Just think like that for the next 10 days, and you will be shocked at how many things you do and say more than three times.”
Sarah Madden 01:16
“The backlog of work is almost infinite, and I'm still not keeping up, even with maxing out my cloud account every single day.”
Anthony Enrico 01:25
“Oh my God, this is how you build a business. This is so cool. This is why you collect data.”
Sarah Madden 03:32
“No, I don't think you understand. I think being in Excel and doing the lookups is way more fun and way more engaging.”
Sarah Madden 05:15
“It's one thing to do some vibe-coding fun things for yourself, but we're tasked to do the job of building things that can be handed to the entire team.”
Anthony Enrico 07:00
“I'll give you a cursor license, go figure it out, see if you can access it in terminal.”
Sarah Madden 09:04
“He's like, stop asking me ask the chat bot.”
Sarah Madden 10:01
“It forced me to go in with no assumptions on how anything worked, because I didn't know how anything worked.”
Sarah Madden 11:11
“If you can read a PDF of contracts, you can read a PDF of event attendees.”
Sarah Madden 11:57
“There were so many things that I've always wanted to do that I just couldn't do because I was held back by bandwidth capacity.”
Anthony Enrico 12:53
“Our job at the core is prioritization, not just for ourselves and for our team but for also the rest of the business.”
Sarah Madden 13:48
“But like any operator I work better with tags and categories than just like a big blank white space.”
Sarah Madden 16:11
“So how do I build a whole set of skills that compensate for my blind spots and my gaps.”
Sarah Madden 18:35
“Turns out, almost every part of your job you could turn into a skill.”
Sarah Madden 20:52
“The answer is almost always yes. And if you're not sure, ask Claude and then it'll be yes.”
Sarah Madden 21:26
“Slack isn't just our messaging app, it's a data lake.”
Sarah Madden 23:14
“That would have taken a data science team of 20 and a whole team of engineers to build before.”
Anthony Enrico 24:48
“We have, you know, a section that's vibes and it counts as 10%. Like those days are over.”
Sarah Madden 25:14
“There is a persona of me in the skill that like interrogates the budget owner, are you sure you need this?”
Sarah Madden 31:54
“In this case, with the renewal and eval, I'm actually shifting the burden to them. I don't know if I've ever really done that in rev ops.”
Sarah Madden 34:12
“AI only surfaces back the evidence for each field.”
Sarah Madden 36:58
“You the human needs to rewrite this in some way. And that's our way of including some human element like some human judge in the loop.”
Sarah Madden 37:28
“If you removed AI, would you still get the same output? And if you still get the same output, then AI has only just made it faster for you.”
Sarah Madden 38:10
“I just got three hours back in the last eight minutes.”
Sarah Madden 42:32
“There could be a handful of enterprise deals buried in those emails.”
Anthony Enrico 43:38
“Any tool that doesn't have an MCP by now is not a tool I'm looking to have much relationship with honestly.”
Sarah Madden 47:01
“If I have to go into your tool to use it, I'm not going to use it. And I'm not going to demo it probably.”
Sarah Madden 47:08
“People follow pain. So whatever your pain is where you're spending time where you can't recreate a workflow with AI, that's probably where you need to hire first.”
Sarah Madden 51:17
“We're basically a data integrity team that's really good at automations and tools.”
Sarah Madden 53:07
“And the go to market data is the messiest data. It's the most difficult data to work with.”
Anthony Enrico 55:16
“We've already done things on hard mode, we might as well just go get the easy stuff too.”
Anthony Enrico 55:30
“The importance of building out loud, I think is even more important, because no one's going to come in and check on you.”
Sarah Madden 56:07
“He said, you should never run a session past 50% context, because that's when Claude and even codex really starts to degrade.”
Sarah Madden 58:48
“I have used wrap and pickup, I use them about 250 times a month, which is pretty crazy.”
Sarah Madden 1:04:12
“I kept getting from a chief of staff, like you don't have any new emails, you're good. And then I realized four hours later, the Gmail MCP had broken.”
Sarah Madden 1:06:10
“This is now a cornerstone of your tech stack. And we have maintenance days, we have code freeze periods for Salesforce, we have blackout dates.”
Sarah Madden 1:12:17
Practical Advice

What should you actually do?

The playbook, split by the seat you sit in.

RevOps Leaders

  • Apply the rule of three: more than three tools in a process means rethink it; anything said or done more than three times should be a skill.
  • Categorise what you have already built before building more — feed the skills to Claude and ask it to group them.
  • Ask the model what it needs from you to work as expected, rather than guessing at engineering hygiene you do not have.
  • Build a tracker for silent connection failures; an agent reporting nothing looks identical to an agent that cannot see anything.
  • Keep sessions under 50% context and build wrap/pickup skills so clearing context costs you nothing.

Revenue Executives

  • Listen for the anxiety in how people describe their work — that is the pain point, not the task they name.
  • Apply the remove-the-AI test: if the output would be the same without it, you sped something up rather than redesigning it.
  • Consider shifting work toward stakeholders where the skill can carry your judgement, not only away from them.
  • Treat agentic infrastructure as a cornerstone system with maintenance days, code freezes and blackout dates.
  • Expect a bifurcation in RevOps — the technical half drifting toward data teams, strategy and forecast management staying.

Sales Leaders

  • Consider surfacing evidence for CRM fields rather than autofilling them, and requiring the rep to rewrite it.
  • Run an end-of-day scan for calls on calendars with no matching opportunity, prompting rather than nagging.
  • Keep a log of every field updated and why, which is what makes the lighter-touch process trustworthy.

Marketing Leaders

  • Stop deleting free-email registrations — enriching them is where the enterprise deals hide.
  • Point the import skill at your documented SOP and naming conventions so campaign creation and member status follow the rules automatically.
  • Run tool renewals yourself through a skill that carries the RevOps interrogation, rather than queueing for RevOps.
AI Takeaways

How AI actually changes GTM

LeanScale's signature read on the AI-in-GTM question this episode wrestles with.

The thesis

This is the most concrete account in the archive of what an AI-native RevOps practice actually looks like day to day. The organising insight is that the interesting work is not the flashy agent but the infrastructure around it — categorising skills, managing context, monitoring connections, and deciding deliberately where the human stays in the loop. Sarah's own framing is that she is not a software engineer, so she asked the model to specify what it needed from her and built skills to cover blind spots she could not name.

Agent & automation ideas

  • A tool evaluation and renewal skill scraping documentation, Slack and spend data, carrying the RevOps interrogation as a persona, and producing a one-pager.
  • A lead-list import skill that cleans a CSV, enriches free email addresses, creates the campaign per documented naming conventions and sets member status.
  • Pipeline ops skills that create opportunities from inbox or calendar and surface field evidence rather than writing values.
  • A missing-opportunity scanner checking calendars for customer calls with no corresponding record.
  • A wrap and pickup pair that documents decisions and resumes a session cleanly after clearing context.
  • A core-tools tracker with a recurring heartbeat so broken connections are visible rather than silent.
  • A customer-health workflow combining shared Slack channels, transcripts and project data to classify accounts.
Operations Takeaways

By function

The same conversation, filtered for RevOps, pipeline/marketing ops, and customer ops.

Revenue Operations

  • .
  • .
  • .
  • .
  • .

Pipeline & Marketing Ops

  • .
  • .
  • .
  • .

Customer Operations

  • .
  • .
  • .
Metrics Mentioned

The numbers, with context

$20,000 quoted, $200/month given
Consultant budget versus tool licence

Two months of contractor support to backfill 500 contract PDFs, replaced by a Cursor licence and a refusal to help.

~450–500 PDFs
Contracts in the backfill

Amendments, renewals and MSAs, all differently named and formatted, that needed to reach Salesforce as subscription line items.

~70
Skills in the go-to-market repo

Built by the RevOps team plus roughly 25–30 go-to-market staff who had each created at least two, before Sarah stopped to categorise them.

~10%
Portion of her job turned into skills

Sarah's estimate of how far she has got applying the rule of three — the ones she is most comfortable with and has most control over.

~20 minutes vs ~3 days
Vendor data bake-off

Running the comparison from the terminal against vendor API keys, versus the vendor running the test and presenting a PDF.

A full day → ~40 minutes, 4 per day
Lead list import

The import, dedupe, campaign creation and member status process a RevOps analyst can now run repeatedly.

~3 hours in 8 minutes
Time recovered on one import

Sarah's reaction as the CSV cleaning, inference and enrichment completed.

~250 times a month
Wrap and pickup usage

Across both work and personal projects, to keep every session on the strongest version of the model.

50%
Context ceiling per session

The threshold a go-to-market engineer told her marks the start of quality degradation, which prompted the status bar.

4 hours
Silent tool outage

How long her chief-of-staff agent reported no new emails while the Gmail connection was actually broken.

Entities

Companies, people & tools mentioned

Auto-extracted and linked into the knowledge graph.

Companies

FERMÀTCommerce

Where Sarah leads RevOps as the first ops hire. AI-native commerce platform selling to e-commerce companies, running a deliberately lean stack and building much of its own tooling — including a lightweight Salesforce-to-SpotDraft integration in place of a CPQ, and a website rebuilt on Claude Code after leaving Webflow. Roughly 25–30 people in go-to-market, essentially a flat organisation.

07:27 · 45:33 · 55:52Company →
BrazeCustomer Engagement

Where Sarah was employee #64, cold-calling as a BDR. The all-hands waterfall chart presented by the CRO is what redirected her career, and the VP of Operations there was her first introduction to the function.

00:16 · 04:59Company →
LeanScaleGTM Operations

Anthony's firm. He describes combining shared client Slack channels, a transcript warehouse and the project management system into a customer-health workflow that classifies accounts, and notes LeanScale pipes a Product Hunt feed into Slack to track new go-to-market tools.

23:48 · 32:45Company →

People

Sarah Maddenguest

Head of Revenue Operations & StrategyProfile →
Anthony Enricohost

Co-founder of LeanScale and host of The LeanScale Podcast.

Co-FounderProfile →

Tools & software

CursorAI Dev Tool

The $200-a-month licence given instead of a $20,000 contractor budget. Sarah did not know what it was, or what terminal meant, and it became the entry point to running her job from the command line.

Website ↗
Claude CodeAI Dev Tool

Where most of the work now happens. Sarah runs it inside a multiplexed terminal rather than the terminal app, with a status bar showing context usage, session spend and the active model.

Website ↗
CodexAI Dev Tool

Used alongside Claude Code, with Sarah sharing in an internal AI enablement channel how she decides between them — explicitly noting she is unsure the reasoning is right but sharing it anyway.

Website ↗
NotionDocumentation

Holds the marketing calendar, campaign SOP and naming conventions, the documentation library, and the RevOps projects and problem-statement databases. Sarah notes scanning it spikes usage cost while writing to it does not.

Website ↗
GranolaMeeting Notes

One of the five core tools the connection tracker monitors, alongside email, calendar, Notion and Slack. Sarah notes she did not have it during the original Cursor conversation and so had no transcript to review afterwards.

Website ↗
RampSpend Management

Pulled automatically by the tool-evaluation skill to compare current vendor spend against alternatives during a renewal or bake-off.

Website ↗
CrossbeamPartnerships

The worked example of the tool renewal skill — the head of marketing kicks it off, and it scrapes Notion and Slack, finds alternatives, interrogates the budget owner through a persona of Sarah, and posts a one-pager to the RevOps channel.

Website ↗
SpotDraftContract Management

FERMÀT's contract management tool. Sarah built a lightweight integration between Salesforce and SpotDraft that serves as their CPQ rather than buying one.

Website ↗
SalesforceCRM

The system the contract backfill was destined for, and the source of the enrichment bake-off dataset. Also the reference point for her closing argument — maintenance days, code freezes and blackout dates should apply to agentic infrastructure too.

Website ↗
SlackTeam Messaging

Reframed as a data lake rather than a messaging app — scraped for AEs complaining about data quality during vendor evaluation, for project context, and as the channel where RevOps releases and sprint launches are posted.

Website ↗
ClayGTM Data / Enrichment

Used extensively in FERMÀT's stack, named among the tools they still buy rather than build.

Website ↗
G2Software Reviews

Checked by the tool-evaluation skill for additional reviews when comparing vendors.

Website ↗
WebflowWebsite Builder

Previously hosted FERMÀT's website, since cut in favour of building the site on Claude Code.

Website ↗
FirefliesAI Note Taker

LeanScale's call recorder, mentioned by Anthony when comparing notes on conversation intelligence and whether video is needed at all.

Website ↗
Methodologies referenced Skill inventory and categorisationProblem statement before projectContext hygiene with wrap and pickupConnection heartbeat monitoring
Related LeanScale Services

Where this shows up in the work

AI for GTM

Agents and workflows that multiply operator throughput — the 'make humans superhuman' thesis, operationalized.

The episode is the most detailed working account of AI-native RevOps practice — skills, infrastructure, context discipline and where the human stays in the loop.

Fractional RevOps

Embedded, field-facing RevOps leadership and execution — the operator model Jimmy describes, delivered as a service.

The prioritisation framing, the tool evaluation process and the pipeline hygiene work are core embedded RevOps responsibilities.

Sales Enablement Build

Stand up a zero-based, sales-academy enablement motion sized to your span of control and sales complexity.

Sarah's forecast that enablement has the next resurgence, and the release-notes and sprint rituals her team runs like a product function.

Frequently Asked Questions

Straight answers

Generated from the conversation, marked up for search and AI extraction.

What are the three buckets every RevOps AI skill falls into?

Surfacing context, process, and infrastructure. Surfacing context covers things you could do yourself more slowly — an email inbox scanner, a scheduled competitor scan. Process means handing a skill an entire end-to-end workflow you then no longer perform. Infrastructure means maintaining the system itself: a healthy terminal, a sensible folder and file structure, monitoring that connections are alive. Sarah Madden derived these by feeding roughly seventy existing skills into Claude and asking it to categorise them, on the basis that operators work better with tags than with blank space.

How do you know when to build a skill?

Three rules, all built around the number three. If completing a process requires going into more than three tools, stop and rethink the workflow — the context switching is the signal. If more than one person on the team follows the process, it has enough breadth to be worth capturing. And if you say or do something more than three times, it should be a skill. Sarah says thinking this way for ten days will shock you with how much of your job repeats, and that almost every part of it could become a skill — she estimates she has built about ten per cent of hers.

How do you identify which problem to solve?

Listen for the anxiety rather than the task. Nobody says they cannot get all their follow-ups out on time; they say they are up late at night working on email and always feel behind. That second sentence is the real pain point. Sarah's other example is someone going into fourteen different tools to assemble the same QBR deck every time — the fact that all fourteen connect to AI and the process is identical each time means it is not a workflow, it is a set of instructions, which means it is a skill.

Should AI fill in CRM fields automatically?

FERMÀT deliberately chose not to. Their Pipeline Ops plugin has Claude surface evidence rather than write values: for a field like economic buyer or champion, it returns several bullet points of moments where someone explicitly said it, written so it cannot be copy-pasted, with a rule preventing that. The human has to rewrite the entry. Sarah is clear that plenty of teams autofill successfully and that this is a choice about readiness — they are not afraid of slowing an agentic process down to keep human judgement in the loop, and they keep a log of every field updated and why.

How do you tell whether you actually redesigned a workflow or just sped it up?

Remove the AI and see whether you would get the same output. If you would, then AI only made it faster — that is not a redesign. A genuine redesign starts from a problem statement, such as needing a clean opportunity pipeline with minimal effort from anyone else and maximum accuracy, and produces capability that did not previously exist. Sarah's example is a skill that runs at the end of the day, checks reps' calendars for customer calls with no matching opportunity in Salesforce, and prompts them — work an analyst used to do by hand.

How should you manage context in long AI sessions?

Sarah was told never to run a session past fifty per cent context, because quality begins to degrade beyond that. Rather than asking the model where she was, she added a status bar showing usage as a visual breadcrumb. She then built a wrap skill that summarises what was decided and built, updates a cache and writes a compact handoff document, and a pickup skill that resumes exactly where the previous session ended. She copies the handoff, clears context, pastes it back, and continues — using the pair roughly 250 times a month.

Why build a tracker for tool connections?

Because a silent failure is indistinguishable from a quiet day. Sarah's chief-of-staff agent kept reporting that she had no new emails, and four hours later she discovered the Gmail connection had broken. She built a core-tools tracker covering email, calendar, Notion, Slack and Granola so she is told when any of them go offline, plus a recurring connection heartbeat. Anthony's immediate response was that he was taking the idea — scheduled routines failing silently is a problem he recognised straight away.

What makes a tool worth evaluating now?

An MCP or a CLI. Sarah's position is that any tool without one by now is not something she is looking to have much of a relationship with, regardless of category — and if she has to go into the tool to use it, she probably will not even take the demo. She applies the same standard personally, having installed MCPs for her fitness tracker and music. She also questions whether conversation intelligence needs video at all: her team uses the transcripts far more than the platform's own insights, and if sentiment can be detected from the transcript, the video may be unnecessary.

How is the RevOps role likely to change?

Sarah expects a bifurcation. The strategy, process, enablement and forecast-management side — which involves working with sellers and sales managers — stays recognisably RevOps. The data, analytics, systems and automation side may drift toward data teams, possibly reporting to a VP of data, since the people already managing sixty to eighty per cent of the tech stack are natural owners of the rest. She also expects enablement to have the next resurgence after RevOps. Her framing to her own team is that they are essentially a data-integrity team that happens to be very good at automations and tools.

Timeline

Jump through the conversation

Click any chapter to open it in the transcript.

00:00Cold open + intro 02:04The waterfall chart that changed her career 04:35From BDR to ops: "this is the job that I want" 07:12A $20K consultant budget, a $200 Cursor licence 10:15Being refused help was the unlock 12:33The infinite RevOps backlog 15:29The three buckets: context, process, infrastructure 19:09The rule of three, and listening for the anxiety 22:28Slack is a data lake 26:16The data enrichment bake-off in 20 minutes 31:28The tool eval skill, and shifting work to stakeholders 34:44Pipeline Ops: evidence, not autofill 40:12Importing a lead list: a day becomes 40 minutes 44:50The tech stack, and no MCP means no relationship 50:42Building a RevOps team now 57:09Infrastructure skills and the 50% context rule 1:01:14Wrap and pickup: 250 times a month 1:06:05The core tools tracker for silent failures 1:08:32The RevOps operating system and releases 1:12:07Treat agentic infrastructure like Salesforce
Full Transcript

The whole conversation

Broken into chapters, searchable, verbatim from the audio. Speakers inferred (not diarized).

Transcript as markdown ↓ Whole episode as markdown ↓
00:00Cold open + intro

0:00 You can, like, drill down into where the data comes from, and then if you have the field of view and someone's shown you how to see the big picture, you can see how that all ladders up.

0:09 Sarah Madden, a.k.a. Smads, is the head of RevOps at Vermont, though, as you'll hear, the title barely covers it.

0:16 She started as employee 64 at Braes, cold-calling accounts as a BDR, until a single waterfall chart in an all-hands convinced her she wanted to spend her career in the data behind the revenue.

0:28 You can hear it. You can hear the anxiety. That's the problem. That's the pain point that we're solving. It's not, "Hey, I can't get all my follow-ups out on time every day." No one says that.

0:36 People say, "I'm up so late at night working on my email, and I feel like I'm always behind." That's the pain point.

0:42 There could be a handful of enterprise deals buried in those emails, so I think those are the type of things where, if you're not using AI, even, let's assume it took the same amount of time.

0:54 Even though you're still leaving money on the table, missed opportunities, missed level of depth to the analysis you're running, you just have such an opportunity to take it to the next level.

1:07 And the last thing that we really anchor on is, if you say something or you do something more than three times, it should be a skill.

1:16 Just think like that for the next 10 days, and you will be shocked at how many things you do and say more than three times. It has been living in my brain rent free for the last two months, and someone told me that.

1:25 Now RevOps is, in my opinion, having the biggest moment of the entire function, because RevOps, I think the backlog of work is almost infinite, and I'm still not keeping up, even with maxing out my cloud account every single day.

1:43 So, I think it's an exciting time, and I think where you've gone, you have some of the most impressive applications of AI in RevOps that I've ever seen.

1:56 Thank you, thank you.

02:04The waterfall chart that changed her career

2:04 Matt, you said the first thing that got you hooked on ops was watching a waterfall chart in and all hands. What was it about that moment that made you walk up afterward and think, "This slide is the job that I want?"

2:19 Yeah, it was pretty much exactly that. I can still tell you where I was sitting in the room of the all hands meeting when I saw this happen.

2:28 And I think what stood out to me at first, when I first saw the chart, I'm like 24 years old, 23, I definitely don't know what a waterfall chart is, and I don't know what these floating bars mean in the middle of the page.

2:37 But as our CRO at Braes, Miles Cleager, who's walking through what each bar meant and how they stack on top of each other, and he stops at expansion and upsell, and he talks us through how that happened and the products, he gets to churn, talks through how that happens, and paints this whole picture.

2:52 And I see just looking up at that stacked bar chart, I'm like, "Oh my God, you have all this information because I put the right closed loss reason on that closed loss op."

3:03 And I thought it was literally the coolest thing that you can drill down into where the data comes from, and then if you have the field of view and someone's shown you how to see the big picture, you can see how that all ladders up to at the time Miles was talking through.

3:17 So I think we were shifting away from having account managers to more CS-based because our only churn was from a product that went out of business.

3:25 And he's explaining this on how it makes sense because account managers have a slightly different focus, and CS has a slightly different focus, and that's what the business needs.

3:32 And I was like, "Oh my God, this is how you build a business. This is so cool. This is why you collect data. This is why it's important to fill in fields, even when they're annoying, so that someone like Miles can stand up here and say, 'This is how we're going to drive the business, and this is how we're going to get the most out of it.'"

3:45 You guys in every single different way.

3:47 And that was very cool because I could see on the floor as an individual contributor, but then at the same time I could see how it all lattered up and how that actually drives the business.

3:56 I can still picture the colors on every bar chart from that waterfall. It was really nice.

4:03 I love that. I have a very similar, I call it a spider bite moment, like when you go from a normal human being to a superhero ops person. And for me it was the first time seeing a growth model and GTM capacity plan.

4:16 And I was like, "Oh my gosh, you can see into the future with this model and everything you would take to go from where we are today to where we want to be."

4:25 And that was the moment I was completely hooked to. So walk me through what happened. You see the chart in a meeting? What was the story after that and how you got into it?

04:35From BDR to ops: "this is the job that I want"

4:35 Well, so at the time when this all hands happened, I was a BDR at Braze. I was already talking to the team there and saying, "I love generally what I'm doing, but I don't want a book of accounts. I look around. I don't want to be an AE. There's got to be something else on GoToMarket.

4:51 I see the slide and I walk up to miles after and I'm like, "This is the job that I want." I don't even know what job it is, but like whose ever job it is and is involved in this, that's what I want.

4:59 And that's when I got introduced to, or spent more time with our VP of Operations, Oliver Bell at Braze at the time. And someone was explaining to me, "I don't think you really want to go into operations. It's a lot of Excel time. You're on the phone with customers. It's really exciting and really engaging."

5:15 And I was like, "No, I don't think you understand. I think being in Excel and doing the lookups is way more fun and way more engaging. Actually, that is the right path for me."

5:23 So that was the first time that I started to see, "Oh, there's actually a whole career around Excel that's not being an accountant."

5:30 And that really started to click things into place. Like my first, honestly, the only class from college that I really have a strong memory of is the class in freshman year when we learned how to do Excel.

5:40 And it was called Omos 121, and I learned all my formulas, all my pivots, and it was so exciting. And then every year after that, I would do the Coachella spreadsheet.

5:49 I would do the March Madness spreadsheet, like anything that requires a system or a spreadsheet. My friends and I just went to Paris two months ago. It's 2026, and I refuse to use any of the expenses splitting softwares because I want to make my own spreadsheet.

6:02 They hated it, but I've always been drawn to that. And if I can be in a system and a spreadsheet in the database, it's so much fun because that's what underlies everything in RevOps.

6:12 Everything we do ends up coming from a system, which is pretty cool.

6:15 Yeah, I always like asking people who are in ops, "What is your personal finance management strategy look like?"

6:21 And it's almost always custom-built, hand-grown, artisanally made spreadsheets because it's like, "Yeah, there's all these apps, but it doesn't do these very nuanced things that I want it to."

6:31 So I just build it myself.

6:35 And now, I think, now RevOps is, in my opinion, having the biggest moment of the entire function because in Go to Market, it's really the team that is being looked at to lead the charge on AI and have a definition of what good looks like and build these things that can actually scale.

7:00 It's one thing to do some vibe-coding fun things for yourself, but we're tasked to do the job of building things that can be handed to the entire team as well as scaling what RevOps is doing.

07:12A $20K consultant budget, a $200 Cursor licence

7:12 So I love one of the stories that you shared in our prep session, too, about you being underwater in your ops role, needing some support, looking for a consultant budget, but then getting a cursor license instead.

7:27 Yeah, now it's funny because now I have the resources and the skills to actually do it. It wasn't so funny at the time, but when I first started at Vermont, again, I was the first hire, so we had some infrastructure built in Salesforce and HubSpot.

7:41 Not a lot. Not a lot of it was repeatable. And we were coming up on some big renewals, and I needed to basically get my arms around our renewal tracking.

7:50 We didn't have a great, great form of renewal tracking, you know, things that slipped or we missed deadlines. So that was the thing that I was really diving into was how can I take this Google Drive folder of like 500 different contracts, PDFs, all in different formats, all with different names of products, like everything that could be different about these PDFs was different.

8:08 They were all named differently in the folder. They were amendments. They were renewals, all different types of things. Some were MSAs. And I had to go through there, figure out exactly the contract line items that we had, and make sure that all of that was reflected in Salesforce on like a subscription, subscriptions object connected to the contract.

8:24 And I'm sitting there thinking, okay, I can do all the Salesforce stuff. I know how to build all this. It's the backfilling that's going to kill me like this is going to take forever. I don't know any other way besides just doing doing the damn thing and like going through every single PDF and hopefully at some point it picks up.

8:39 We've all been there. And that's when I said to my co founder, I was like, Listen, if you give me contractor support, I can probably get this entire thing done in under two months, I can do this part of building and then I can outsource this part, you know, the PDF part.

8:54 And my co founder said, Well, why can't cursor do that? And I was like, I don't know what cursor is. So I don't know why cursor can't do that. And then we get that we get into this back and forth. He's like, I'm going to give you a cursor license first.

9:04 It's like $200 a month, you know, because of course, I quoted something around like 20k for two months to bring on someone to help me with this work. And he says, I'll give you a cursor license, go figure it out, see if you can access it in terminal.

9:16 And this whole conversation, I'm like, what's cursor? What's terminal? I don't know what you are even saying to me. So I'm not really sure how to respond. And of course, I didn't have granola at the time. So I didn't have a transcript that I could run through Claude and like get some help on what just happened in the conversation.

9:30 So he gives me a cursor license, I did some googling, saw a lot of words that I don't really understand. And he got on the phone with me like the next day or something. And we sit down and, you know, he's showing me I'm not even in applications where the whole thing is black, right?

9:43 Like chrome windows, right in settings panels, this whole thing is new to me. And we're sitting there and he shows me like, look, you see down there where the cursor is blinking? That's the text box. It's the same thing as in chat GPT or is in Claude.

9:54 So all of a sudden, I'm realizing like, oh, this is kind of the same thing. It's just a different layout. And then I keep asking questions like, oh, well, how do I do this? How do this?

10:01 And he's like, stop asking me ask the chat bot. So eventually I was like, okay, I get it. We keep moving for about an hour. And then I realized, okay, there's a lot here. Let's let me let this settle. Let's regroup in a day in a day or two.

10:15Being refused help was the unlock

10:15 And we set time for Friday. It was like Friday for an hour. And I'm getting ready five minutes before, of course, I want to prep before my co founder gets on the phone. And I'm sitting there. I'm like, okay, I'm gonna ask him this question. I'm gonna ask him this question.

10:25 And I'm like, wait, he's just gonna tell me to ask the chat bot. So why do I I'm not gonna wait for him to get on the phone. I'm just gonna start asking the chat bot and instantly cursor and I start going back and forth. I don't even know what model I still don't even know what model I was on in cursor at this point, I had one folder on my desktop that had every single project I'd ever worked on.

10:42 It was a mess. And I am going back and forth with cursor. And then I eventually slack my co founder. I'm like, Shreyas, you know what? I don't think I need you because I don't know what are you gonna do just sit here, watch me type to this chat bot. Like, I think I figured it out. I get the idea I can I can do it.

10:55 And he goes, that's good, because I'm getting on a flight. I wasn't going to join anyway. And I was like, okay, I get it. I get how much we can do ourselves now. And that like both of those things like getting the cursor license, and then being told, yeah, I'm actually not going to help you because you don't need help.

11:11 You sit down and you can figure it out was the perfect forcing function for me, because it forced me to go in with no assumptions on how anything worked, because I didn't know how anything worked.

11:21 And I asked every question like really honestly, and I didn't have any crutches to turn around and be like, well, what do you think? How should I say this? I just had to talk to the computer, which was crazy for a few months.

11:32 And then I realized like, what could be built? I gate I was asking cursor like, okay, here, I have this project, I've got this PDF folder. Here's how I'm going to go through it. And cursor said, Well, why don't you just give me the link to the Google Drive folder, dude, I can still picture this on my computer screen.

11:45 And I was like, just the link. And I give it to cursor and it starts telling me, you know, you've got 450 contracts in here, blah, blah, blah, blah, blah, blah. It wasn't perfect. Of course, this was almost two years ago now.

11:57 But it was the most impressive thing that I'd ever seen. And I was like, oh my god, if you can read a PDF of contracts, you can read a PDF of event attendees. And you can read a PDF of all these other things of sales collateral, all of that is possible.

12:10 And as soon as I took that one, keep calling it a non technical use case, because I will like to stay thinking that I am non technical.

12:17 As soon as I realized, oh, there's a pattern here in terms of like what I can give it and what I can get out and then how I can act on it. That was the initial foundation that I really started building on when it came to what is AI capable of regardless of what tool I'm using.

12:30 So that was a really big turning moment. Yeah.

12:33The infinite RevOps backlog

12:33 Yeah, and I think you're gonna have to reevaluate that definition of technical I think a lot of us are. But I think there's something so dangerous about someone in rev ops just getting completely weaponized with AI, and realizing, I think the level of curiosity just expand so much because

12:53 there were so many things that I've always wanted to do that I just couldn't do because I was held back by bandwidth capacity. It's like, oh my gosh, I want to do this analysis, but I need a whole like date engineering team in order to do that or I want to be able to model this out in order to build a model that's going to be like eight hours in Excel.

13:10 I'm not going to spend my entire day doing that's not worth doing it. But I had a list of 10 million things that I would love to be able to do that I just could never get to.

13:20 So that's why at least in rev ops, I know there's some jobs that are absolutely going to be impacted by AI. But rev ops, I think the backlog of work is almost infinite, and I'm still not keeping up even with maxing out my cloud account every single day.

13:35 So I think it's an exciting time. And I think where you've gone, you have some of the most impressive applications of AI in rev ops that I've ever seen.

13:48 Thank you. Thank you. Yeah, I think, I think back to your point on, you know, bandwidth and stuff like our job at the core is prioritization, not just for ourselves and for our team but for also the rest of the business, which means we have to be the most ruthless

14:01 prioritizers in the business, which means there's always going to be stuff on our backlog, there's always going to be stuff that we don't don't do. And then the other flip side of our job is pattern recognition and seeing further out than right now.

14:14 So there's always going to be things on the list and to your point like if we can start coming up with ways to get 10% of my time back so that I can explore some of these things that are further for more forward looking or have been on my backlog because I'm not sure if I can quantify it being a priority.

14:29 Now I can quantify and be sure it's not a priority, or it will flip my mind be like oh my god I did not realize how much this is blocking. Now I can reprioritize, which is a gift, like being able to dip into all the things we've always wanted to do, and be able to prioritize

14:45 better so that we're working on better things, I think is a huge benefit of all of the AI.

14:50 But yeah, thank you. Thank you for your comment about like good applications. Go ahead. Fully agreed, fully agreed. Well, I know you prepared some demos today or walking through some of the things you've built.

15:03 Man I think anybody in RevOps is just going to love this. I hope so. I hope so and I have prompts on GitHub for like the skills that you know you can't really replicate and share with someone else I've got prompts that we can share.

15:15 We can download themselves. I thought I'd go through a few real like RevOps go to market workflows that we've done or maybe I'll start by explaining kind of how we think about creating skills when to create skills, what buckets we're categorizing them in.

15:29The three buckets: context, process, infrastructure

15:29 Then I think that will make sense on like the RevOps skills that we've really built. And then we have a whole other set of what we call infrastructure skills that are all about helping us maintain this new system that we now have similarly to how you have like maintenance days for Salesforce or upkeep for your key

15:45 and core platforms and systems. We're treating all of our AI platforms the same way as that we're already thinking, obviously we're thinking about tracking, but we're also thinking about how do we maintain the system for each person and for each of us over time.

15:57 I can go through, let me pull up. So let's talk through a few of the different patterns that we found at Vermont. I'll start by saying, certainly none of this is exhaustive, and this is just the patterns that we've seen in our work and I'm sure we're missing things so I'd love

16:11 to add to our buckets. But like any operator I work better with tags and categories than just like a big blank white space. So the first thing after we got I'd say like maybe the RevOps team had done about a dozen skills ourselves.

16:24 We didn't have any plugins yet. It was just individual skills, individual MCPs connected. And I'd say the rest of the team at Vermont, I'll talk a little bit more about our building approach too.

16:35 But the rest of the team, let's say across our 25 to 30 people on go to market, everyone had created at least two skills, probably. They've used code a little bit they've used codecs.

16:46 We're probably working most of their time in co work, and then some other people are working in code and codecs. So we're already a pretty advanced group, or a familiar and a comfortable group, we're all at different levels.

16:57 We have a nice baseline but we're all at different levels. So at this point when me and my team said okay we need to stop and look at everything that we've built, where's it coming from, where does it live, is it in the same GitHub repo like just organization across the board.

17:09 The first thing we did was we took all the skills, probably about 70 skills in our go to market repo, fed that right into Claude and I said, tell me what's in here.

17:18 Help me categorize them, move them into groups, explain to me what they mean, do different ones have different permission sets like just understanding what we had already done.

17:26 And a few patterns came out for us so when I look at the skills I group them into three different buckets in terms of what they can do for us, or like what their purpose is.

17:37 So there's one set of skills, that's about surfacing context, right like your email inbox scanner, it surfaces information to you that you probably would have done otherwise.

17:46 You maybe have a scheduled task that scans the market or scans for competitors, all of that is surfacing information that if you didn't have AI, you could probably go out and do it yourself.

17:55 It may not be the same quality or this as exhaustive but you could do it yourself. That's really helpful because it gives us time back, it gives us information in the right way and exactly the right way that you consume it better than how I consume it.

18:06 So that's surfacing context. The second one is where we get into process, and that's where we actually give a skill an entire end to end process or a couple of combination of skills, a whole process that the skill is now expected to do.

18:18 And I do not have to do that process at all. Those are some of the rev up skills I'm going to talk about later.

18:23 And then the third bucket is infrastructure right now. So, how do we maintain a healthy terminal, how do we maintain a hygienic folder and file structure, I don't know I'm not a software engineer, I have no idea.

18:35 I just had one massive folder that was titled the name of the one project I ever worked on and everything was in there for like 12 months. So, how do we as not being a senior software engineer as not being technical, I don't know what I don't know.

18:49 So how do I build a whole set of skills that compensate for my blind spots and my gaps. So that's infrastructure. That's my favorite bucket at the moment because of how much I've been in code and codecs but I won't, I won't spend too much time on there.

19:02 There's other a few other things that kind of go into how and when we build skills that was very clear coming out of this analysis.

19:09The rule of three, and listening for the anxiety

19:09 We've got a few three rules that's centered around the number three. One is if you are doing a process that goes for anyone across the team if you're doing a process that requires you to go into more than three tools, or three for your more.

19:22 That means we need to rethink the workflow. I don't know I don't know what the problem is I don't know if there is a problem, but we need to stop and say, you're clicking too much you're going into too many windows your context switching.

19:31 This does not have to be this way, help me understand what's trying to what's trying to happen. When I'm sitting down with our, you know, F D E's or F D P M's our account managers, and they're talking to me on how they're like you can hear it you can hear the anxiety.

19:45 That's the problem. That's the pain point that we're solving. It's not, hey, I can't get all my follow ups out on time every day. No one says that people say I'm up so late at night working on my email and I feel like I'm always behind.

19:56 That's the pain point that we have to follow and that's the one that we usually get to Claude. So we'll say, okay, Caitlin has to go into 14 different tools to pull something for a QPR deck, and she's going to pull the same thing every time.

20:08 Well if every one of those tools somehow connects into AI, and then she does the same thing every time that's just a skill, that's just instructions. So can we turn that whole thing into a skill, or can we take the biggest time suck and turn that into a skill.

20:20 So that's our approach we're looking for things processes that someone more than one person on the team follows that way we can get as much breath as possible.

20:28 We have to go into more than three tools to complete the process or to get the full value or to understand it. And then the last thing that we really anchor on is if you say something or you do something more than three times, it should be a skill.

20:43 Just think like that for the next 10 days and you will be shocked at how many things you do and say more than three times, it has like been living in my brain rent free for the last two months and someone told me that.

20:52 Turns out, almost every part of your job you could turn into a skill. I've done about 10% of them, the ones that were most comfortable with, and that we have the most control over.

21:02 But as soon as you start seeing that pattern, you realize, oh, running a tool evaluation is something I do more than three times, doing a tool renewal and the business case and the memo and getting stakeholder buy in.

21:14 There's actually an SOP that I follow, even if it's just in my head.

21:18 And then you start to realize, oh my god, if I just wrote that whole thing out, could Claude do it end to end.

21:23 Yes. The answer is yes.

21:26 The answer is almost always yes. And if you're not sure, ask Claude and then it'll be yes. Exactly.

21:32 The answer is try. The answer is ask first.

21:36 No, I think that's a really good framework to think about because if you're going throughout your day and just being thoughtful about logging, hey, these are the things that I'm doing, the actions I'm taking.

21:47 I really like that threshold of if you're doing something three times or more, then it should be a skill and also navigating multiple applications.

21:59 I think that's a big one, too, because I think one blocker, at least as of now in July of 2026, a lot of people have is like, oh, I didn't realize I could connect all these MCPs or I could push data through with a CLI and I can bring it all in with one skill.

22:16 So I think those are really good benchmarks to start saying, OK, start making the list of the things that you want to build and make. And if you don't know how to do it, just ask Claude.

22:28Slack is a data lake

22:28 Just ask Claude. Just start. Just say, Claude, go through my email. What's something I've said more than three times. What's something I've done? Go through Slack.

22:34 I think that's another thing, too. I don't know if we talked about this in our prep session, but someone I was speaking to this woman, Piper Martz, at one of our workshops over the weekend, one of the cloud workshops that we host.

22:45 Piper said something about systems thinking and how systems thinking. I mean, we've all talked about this in the last few months.

22:50 Systems thinking is going to be a core element, especially for like non ops and non operator people learning how to think in systems.

22:56 I mean, we were born thinking in systems. So now being able to take frameworks like this or rules of three or whatever it is and share it with other people and then be able to say, like, this is how we look for patterns.

23:07 We look to find what's the trigger action or what's the outcome or what's the decision we're making. Those can all surface patterns.

23:14 And then once you get into that zone and you realize, oh, Slack isn't just our messaging app, it's a data lake.

23:21 There's actually all of this qualitative data that I never would have had access to if I couldn't just scrape Slack.

23:29 So once you start seeing your tools, which are your systems, you see notion, you see your call recorder, your email and drive, all of a sudden you're like, oh, my God, I see how this can all work together.

23:38 And that's when people really start to light up, which is my favorite part.

23:42 Can I share one that we use that it highlights the use of Slack.

23:48 So we have tons of engagements and we're working with a number of fast growing B2B SAS and AI companies.

23:56 And we have a shared Slack channel with all of them.

23:59 And that's where most of the work is happening.

24:01 So we have a shared Slack channel.

24:03 Then we have all the transcripts.

24:05 So we built a more like a data warehouse to have all the transcripts that we were going to organize them by client, things like that, instead of the automations to get the pipeline of transcripts moving.

24:17 And then we have our PM system.

24:18 So all the tasks and projects that we're doing for our customers.

24:23 Never have I been able to understand the health and sentiment across our customer base until I connected all of our Slack channels, all the transcripts.

24:34 And then our entire project management system all into one customer health agent workflow system that goes, looks at everything and then categorizes which ones are in good health, average health or poor health.

24:48 That would have taken a data science team of 20 and a whole team of engineers to build before.

24:55 And now that one took me like an hour to find.

25:01 I mean, how many times have we done and calculated customer health scores till I'm blue in the face and as the, you know, the caveat and the footnote is always like, well, there's also qualitative stuff that we can't really account for.

25:13 And we don't really know how to include here.

25:14 And we have, you know, a section that's vibes and it counts as 10%.

25:17 Like those days are over.

25:19 How cool is that for customer health?

25:22 I almost always just relied on CSM sentiment because it was the, it was like, that's as close as we're going to get because every product related thing had its caveat.

25:33 It's like, oh, they're still ranting.

25:34 They're super happy.

25:35 They did testimony with us, but they're not using the product.

25:37 It's like, okay, fine.

25:38 Forget it.

25:39 Just tell me what's going on.

25:42 But now we can actually see all the relevant data and do something useful with it.

25:48 It's really cool.

25:49 And the analysis that we can now do on like, you know, when one comment is mentioned in the sales cycle or this piece of content happens after this kind of demo with this stakeholder, it materially moves a deal faster.

26:00 Like, let's be real.

26:01 No way I was ever going to find an insight like that from a pivot table that was not ever going to happen.

26:05 And I probably wasn't going to work hard enough to find it.

26:07 Now we can.

26:08 And that's really cool.

26:09 And now we can make a custom piece of collateral for every customer who walks through if we just make it a skill.

26:14 It's cool.

26:15 Yeah.

26:16The data enrichment bake-off in 20 minutes

26:16 Let me walk you through some of our RevOps team skills and that way we can talk a little bit more about like the process.

26:24 So I want to give a little bit of context.

26:26 This is like a good RevOps story on how every single one of these true process related skills RevOps came about.

26:33 We were doing an evaluation on data enrichment providers.

26:37 Right.

26:38 And we were doing a data test to compare against the data we had in Salesforce, the data we had from our existing providers against potential ones that we were going to move to.

26:47 You know, the onus was on me.

26:50 I had to send an email to all of the people that we were evaluating saying here's our data set from Salesforce, 500 accounts, 500 contacts.

26:57 And when we've had to do that before, I have to go through and say like, OK, this percentage is going to be like dirty and bad data.

27:02 This percentage is going to be good data.

27:04 I'm going to do 60 percent are number one ICP, 40 percent are second ICP.

27:09 I'm going to do 10 percent international.

27:10 Like you try and come up with all these different weights and then eventually you just say, you know what, I'm just going to randomize this because this is so much work to get a really good data test manually.

27:19 And at some point you can't spend more than like two hours on that spreadsheet.

27:22 So I was getting I was preparing myself.

27:24 I was like, God damn, I'm going to have to go in and do this.

27:26 And then I thought, wait a second.

27:28 I think I just did something with code where it looked at a CSV.

27:32 So I thought, OK, if code can look at a CSV and I know it can make files, so I know it can make a CSV and I just installed the Salesforce CLI, I don't even know what that does yet.

27:41 Is there a way that code could give me a spreadsheet so that I don't even have to go into Salesforce?

27:46 That would be the dream.

27:47 So I took that whole question and I said that to Claude.

27:50 And of course, you know, before you can even like process the response coming back, it's like, oh, sure, Smads, no problem.

27:56 Let me pull these account owners. And it started repeating back to me all of that stuff that I had in my head about ICP Mix, Dirty Clean Data, International US, whatever it was.

28:06 And all of a sudden I was like, oh, my God, this is crazy.

28:10 I can't believe this is possible.

28:12 And again, you know, you see the patterns and I'm like, wait, this is basically my whole job, right, being in tools or being in a spreadsheet and then connecting things elsewhere.

28:20 I can do that from terminal.

28:21 Oh, my God.

28:22 So then we refine the spreadsheet.

28:23 We get it better.

28:24 I get like contacts, accounts and leads.

28:25 We did everything.

28:26 I really fine tuned like the percentages and the weights.

28:30 I felt so good about it.

28:31 I double checked with my pivot table.

28:33 Everything looked good.

28:34 Sent it back to a few of the data providers.

28:37 One of them said, oh, actually, you can run the data test yourself.

28:40 You can upload it into the platform.

28:42 I was like, great, I can do it myself.

28:43 Love it.

28:44 I'll just do my V lookups, compare, you know, the data accuracy, the fill rate, all that stuff.

28:49 I log into the platform and I see they have an API key.

28:52 I was like, oh, wait, I can put an API key.

28:56 Not pasted in a terminal, but there's something with API keys in terminal.

29:00 And then I went back to terminal or went back to cursor at this point and I was like, if

29:03 I gave you the API key, that means you could just pull the data and then you could match

29:07 it against the CSV and then you could do the analysis, right?

29:10 I was like, yeah, of course, just give me the API key, you idiot.

29:14 So then I go through this.

29:16 I'm like, okay, drop it in.

29:18 And within like 15 minutes, I had done the data test.

29:21 It was crazy.

29:22 The whole thing was done.

29:23 And then I said, okay, the data quality isn't that great or it's better, whatever it was.

29:28 Wait, you also have access to email.

29:30 You have access to ramps.

29:32 You know how much we're spending on our current tool.

29:34 You have access to Notion.

29:35 So you know in our documentation library what docs rely on this kind of data.

29:40 You have access to Slack.

29:41 So if I wanted to scrape Slack and see somewhere where the AEs are saying this data sucks,

29:47 I could compliment that onto my analysis and say actually there's a lot of evidence that

29:51 people don't want this provider for whatever reason.

29:53 And we can also pull Salesforce to actually pull and see, you know, are these numbers

29:58 going to convert?

29:59 And when we do occult calls, it can convert to the right number.

30:01 Are we getting the right data?

30:03 The whole thing was done in like 20 minutes.

30:05 It was insane.

30:06 And then Claude put together this like PDF overview of the entire thing.

30:09 And you've sat through these data test reviews with providers before.

30:13 It takes three days to run the test.

30:14 No fault of their own.

30:16 They run the test.

30:17 They come back.

30:18 They walk you through the PDF.

30:19 And I'm like, yeah, yeah, yeah, yeah, just give me the headline number.

30:22 Anyway, I sit there with Claude.

30:24 We run that one.

30:25 And then I'm like, wait, we have three other to evaluate.

30:26 And we just go through the same process.

30:28 And we check RAMP.

30:29 We compare it against, you know, the current provider spend.

30:32 I'm like, go check G2.

30:33 See if there are more reviews.

30:34 And I'm just trying to think of like all of these tools that I have access to, how could

30:37 they enrich my decision on which provider to move forward with without me doing any

30:43 work?

30:44 I'm like, okay, I'm going to do this.

30:46 I'm going to do this.

30:47 And I'm like, okay, I'm going to do this.

30:48 And I'm like, okay, I'm going to do this.

30:49 And I'm like, okay, I'm going to do this.

30:50 And I'm like, okay, I'm going to do this.

30:51 And I'm like, okay, I'm going to do this.

30:52 And I'm like, okay, I'm going to do this.

30:53 And I'm like, okay, I'm going to do this.

30:54 And I'm like, okay, I'm going to do this.

30:55 And I'm like, okay, I'm going to do this.

30:56 And I'm like, okay, I'm going to do this.

30:57 And I'm like, okay, I'm going to do this.

30:58 And I'm like, okay, I'm going to do this.

30:59 And I'm like, okay, I'm going to do this.

31:17 And I'm like, okay, I'm going to do this.

31:18 And I'm like, okay, I'm going to do this.

31:19 And I'm like, okay, I'm going to do this.

31:20 And I'm like, okay, I'm going to do this.

31:21 And I'm like, okay, I'm going to do this.

31:22 And I'm like, okay, I'm going to do this.

31:23 And I'm like, okay, I'm going to do this.

31:24 And I'm like, okay, I'm going to do this.

31:25 And I'm like, okay, I'm going to do this.

31:26 And I'm like, okay, I'm going to do this.

31:27 And I'm like, okay, I'm going to do this.

31:28The tool eval skill, and shifting work to stakeholders

31:28 eval skill. So that now when our head of marketing, let's say

31:32 cross beam, for example, our renewal of cross beam is coming

31:34 up, we take the tool eval skill, the head of marketing kicks it

31:39 off. So I don't even have to do it anymore. She'll kick it off

31:42 and say, Okay, we're about to renew cross beam, or should we

31:44 renew cross beam, go through it, it scrapes cross beam notion

31:48 slack, all of these data sources, puts it together, finds a

31:51 best next alternative and then puts together, well, before it

31:54 puts together the notion doc, there is a persona of me in the

31:57 skill that like interrogates the budget owner, are you sure you

32:00 need this? Is there another way you could do this just like, as

32:03 if I were in the room, and then the final copy they create or

32:06 Claude creates a one pager from notion and then that gets pushed

32:10 to the rev ops channel. And it says smads and Evelyn who's our

32:13 head of finance, there's a new tool renewal proposal or tool

32:17 eval proposal, take a look and then we get feedback. And then

32:20 from there, that's that's now how we process our tool renewals

32:22 and evals is crazy.

32:25 I think what's also really relevant for this one in

32:27 particular is the go to market tech landscape is just exploding

32:32 with new tools every single day. And there are some fantastic

32:37 point solutions that would fit your exact process. But it's

32:42 very difficult to keep up. I mean, this is literally all

32:45 lean scale does every single day, we have hundreds and

32:49 hundreds of tools that we've managed. And it's still hard for

32:52 us to keep up. And every single day on product hunt, we have a

32:56 product hunt feed into slack, like letting us know any new go

33:00 to market tools. Yes, absolutely. And every single

33:04 engagement we do, we run into new tools that we hadn't heard

33:06 of before. So when you're sitting in one organization, and

33:13 you have one tool to renew a couple tools to renew, I mean,

33:17 it's almost impossible to do without AI powering it. So you

33:21 can know, here's all the solutions. Here's all of our

33:24 processes and things that we actually need and things that

33:26 matter to us. Are there certain regions of data you need or

33:29 whatever? And that's huge.

33:32 Yeah, it's, it's a big, it's a big lift off of us. And I think

33:36 what's also interesting, you know, I like grew up in rev ops,

33:40 being under the assumption that like, we are here to make it so

33:44 effing easy for the sellers to sell anything they need that

33:47 does not involve them selling and talking to a customer, if

33:51 they say they need it so that they can bring in more customers

33:54 and support more customers and bring in more business smash,

33:56 you better go do it. Right. And in a good in a good way, in a

33:58 way that it really pushed me to think about how easy can I make

34:01 it for my stakeholders, whether I'm asking you to respond to a

34:03 slack message, or I'm asking you to update your pipeline, how

34:05 easy can I make it. And now we have these tools that can help

34:08 make it easier or even take the burden off of them. Or in this

34:12 case, with the renewal and eval, I'm actually shifting the burden

34:14 to them. I don't know if I've ever really done that in rev ops

34:17 where I say, No, I'm not going to do this work, you're going to

34:19 do this work. So that's like an interesting thing. I haven't

34:23 decided how much I like yet. So far, it's been useful. We've run

34:27 three renewals through it. And we've run, I think maybe like

34:30 three or four brand new tools through it. And it's worked

34:33 every time. And we're really happy with it. So

34:37 Amazing. What else? What else are you deploying into the field

34:41 that's making your ops team do what you can do?

34:44Pipeline Ops: evidence, not autofill

34:44 We've got Okay, so I want to talk about the pipeline ops

34:48 plugin that we just released. And there are a few skills in

34:51 there. And then I want to

34:52 hang on, can I pause on there? You're the only other person I've

34:56 heard use the term pipeline ops. Oh, really use it. We use it

35:00 here at least here all the time. And as like, it's kind of like

35:02 marketing ops for like a subset of more like specific and some of

35:05 it leads into rev ops. But anyway, just I like I love the

35:09 term.

35:10 I like the term and I think it's really clear. It's better for me

35:14 to say who's managing pipeline ops. It's not as clunky as like

35:17 who's the op owner who's responsible for updating this.

35:20 And you know, in most of our orgs, it's not just the sellers

35:22 who own an opportunity, it might be ams or ftpms or account

35:27 managers, whoever it is. So I don't like to just group and say

35:29 like the seller the account manager or the ops owner. So we

35:33 really defer to saying pipeline ops is the task like the job

35:37 that needs to be done. But yeah, I'm glad you like it too. So we

35:41 this pipeline ops skill, well, it's a series of skill is a

35:44 plugin that we just released two or three weeks ago to the sales

35:47 team. And the impetus for this all started with obviously the

35:52 AEs, we have a very lean team, we have a really healthy

35:55 pipeline. And obviously, that means we're not going to have

35:57 the healthiest or the most hygienic pipeline, it's going to

36:00 get stale because that's a lot to maintain for our AEs. We do

36:03 have some level of like fields being updated from our internal

36:06 call recorder and like other tools. But you still need to get

36:09 opportunities into the pipeline. We have some automated ways to

36:12 do that. But our process doesn't always follow the same thing

36:14 every time. And more often than not someone needs to create an

36:17 opportunity like from their inbox or from their calendar, not

36:19 through automation. So to do this to make this as easy as

36:23 possible, we came up with a series of skills starting with

36:27 new opportunity created. That's one skill and I'll walk you

36:31 through what it does. But it's I mean, you can figure it out new

36:33 opportunity created, then we have a separate skill that picks

36:35 up once the off has been created. And that actually changes the

36:39 opportunity stage. And within that this is my favorite part

36:43 within the actual pipeline movement skill, we of course

36:47 connect to the internal call recorder, your email, your

36:49 calendar, like everywhere that you would get the information

36:51 from in order to update your pipeline. And instead of AI

36:55 actually writing any of those fields or making changes to any

36:58 of those fields, AI only surfaces back the evidence for

37:02 each field. So let's take like the economic buyer or the

37:04 champion or legal process, whatever, Claude isn't going to

37:07 surface and say, Hey, Anthony, here's what I think you should

37:10 put in that free text field. Yes or no, and I'll push it over.

37:14 Claude will say back, Hey, here are four bullet points of times

37:17 when someone's explicitly quoted that, you know, Caitlin is the

37:21 economic buyer. And it's written in a way that like you can't

37:24 really copy and paste it. And also we have a rule in that you

37:26 can't copy and paste and Claude's not going to push it

37:28 over. You the human needs to rewrite this in some way. And

37:32 that's our way of including some human element like some human

37:34 judge in the loop. A lot of people have a lot of success

37:37 with like updating all of their fields and Salesforce on

37:39 opportunities. That's awesome. We're not ready for that. We

37:42 didn't want to do it. We want more human in the loop and we're

37:45 not really afraid of slowing down a whole agentic process to

37:49 make sure that we have people checking things. It's also

37:52 helpful for the AEs so they stay on top of you know, all of their

37:55 opportunities that they're running. That's one of my

37:57 favorite because it really shows that we're not just relying on

38:03 AI to do something for us. Like I think I was reading an article

38:06 recently, I'll have to find it about how you can take your AI

38:10 work and your work that you're doing with AI, if you removed AI,

38:13 would you still get the same output? And if you still get the

38:17 same output, then AI has only just made it faster for you.

38:20 That's not really redesigning your workflow. Redesigning your

38:23 workflow is sitting down saying, Okay, my problem statement is

38:26 that I need a really clean opportunity pipeline with as

38:28 little work on someone else. And as accurate as possible. I can

38:32 get accuracy from tools, I can use AI and skills so that it's a

38:36 low lift. And then I know I can trust it because I have a log of

38:39 every field that was updated and why as an admin. Now I have a

38:43 whole new set of like process that I can take, I can say

38:46 actually, I can make a skill that triggers at the end of the

38:49 day. And it goes through their calendar and it checks to see

38:51 if there are calls with people that don't have an opportunity

38:54 in Salesforce. And it will prompt them and say, Hey, should

38:56 you put this opportunity in Salesforce? Let me help you

38:59 update it. And let's check the call to make sure that you're

39:01 putting it in the right stage. And you're matching against our

39:04 entrance and exit criteria. Easy.

39:08 Yeah, and I do like what you said about between efficiency and

39:12 effectiveness, focus on effectiveness, how do you do

39:15 more? How do you make it higher quality? A lot of the things

39:19 that we're doing, they don't necessarily have to be done

39:21 faster. But there are a lot of opportunities to do things

39:25 better. And I think that opens up a ton.

39:28 Yeah, yeah. And then that's how we're doing the pipeline ops, we

39:32 also added kind of on top of the pipeline theme, we added a

39:36 missing opportunity scanner. So you know, sometimes

39:39 opportunities don't get into the pipeline when they should. And

39:42 once upon a time, I used to be a sales ops analyst, and I would

39:45 literally look through people's sellers calendars, and then

39:47 follow up and say, Hey, you haven't logged the meeting in

39:50 Salesforce, can you do it? This was before we were using gong

39:53 and everything else. Now we don't have to do that. And we

39:58 have all these tools that can make it so much easier. And we

40:00 don't have to be involved. I don't have to be the nagger

40:02 anymore. Not I hope I was in the first place. But now we have

40:05 these tools that can do things for us. And it just completely

40:07 changes the way that we should even approach a problem

40:09 statement and like who should complete it. So there's a lot of

40:12Importing a lead list: a day becomes 40 minutes

40:12 a lot of good stuff. Let me walk you through this import lead

40:15 list one because I think this is like the best biggest pain of

40:20 every rev ops person's experience. I have shared this

40:23 with a few rev ops people. And when I walk them through it,

40:25 their eyes light up. And I'm like, don't worry, I will send it

40:27 to you. So anybody who wants it came out after we did the tool

40:33 renewal and evals and realizing like, okay, the all of these

40:35 steps, all these tools, these are just instructions, we can put

40:37 them in a skill. A few days later, I'm going through the

40:41 typical importing a lead list right from a webinar. We have

40:44 one great one webinar partner in particular, they always have

40:47 like 500 leads, they allow free email addresses, I don't like it,

40:51 you know, there's always a misspelling in the email,

40:53 someone's name, whatever. Of course, we usually go through

40:56 that I try to clean it up, I command find hotmail and iCloud

40:59 and all that stuff, and try to clean it up so that I can match

41:02 it to do bit get into Salesforce get into the campaign and set

41:05 the campaign member status. That's the whole process that

41:08 we're trying to do. Of course, I have to go into data loader, the

41:11 CSV, I don't know, maybe I'll check something in the data

41:14 enrichment tool, I'll go into Salesforce. And I'll get

41:17 interrupted five times that it will end up taking me the entire

41:19 day to actually get through this. And there are so many spots

41:22 along this process, we're like, if the member status isn't right,

41:24 or isn't marked as responded, that throws off lifecycle

41:27 reporting. So I have to be really on top of it, or whoever

41:31 on my team is doing it, we need to be really in sync, and we

41:33 really have to have the process written down. So this is just a

41:36 high problem area of importing lead lists, given the way that

41:39 we were set up about a year ago. At this point, I'm like, okay,

41:43 we just did the data enrichment thing with the CSV, we've got

41:46 to be able to do something with the import lead list, I download

41:49 a Google Sheet, save it as a CSV, it's just sitting in my

41:51 downloads folder, I'm not even going to move it. And I say to

41:54 Claude, check out that list. That's in my downloads folder,

41:58 this is an event or a webinar we just had, you can go to notion

42:01 to the marketing calendar to find all the details on that

42:03 event. You know, take a look and let me know what what you see.

42:07 And then I said, you know, it's very impressive. It says there

42:10 are three people who spelled dot com instead of dot com. Someone

42:14 you know, didn't put their last name, but I can infer it from

42:15 their email, like all of the things that a human I would have

42:17 done, instantly done for me, and everything's filled in the CSV.

42:22 Then we talked through what are we gonna do about the free email

42:24 accounts, I said, Can we use the lucha API key again, all of a

42:27 sudden, they're all enriched to find if there's a work email,

42:29 everybody has a work email, and titles filled in like all of a

42:32 sudden, I'm like, Oh my god, I just got three hours back in the

42:36 last eight minutes. That's crazy. And then the process just

42:40 keeps building on itself. And I realized, oh my god, you can

42:43 make a campaign in Salesforce. Actually, if you go to the

42:45 documentation library in notion, you will see the exact SOP and

42:48 the naming convention for our campaigns. You'll also find the

42:52 Google Sheet that has a map of our campaign member statuses,

42:55 and how to add them. All of a sudden, the campaigns created,

42:58 and the campaign member status is created. And then the only

43:00 thing left to do is to get into Salesforce. So we run D dupe, we

43:03 match, we create, everything kicks off, and the whole process

43:07 has taken, it now takes about 40 minutes. And we have a revops

43:10 analyst and he can run like up to four in a day. It's, it's

43:14 crazy. I don't know how else to say it. It's wild.

43:16 Well, and, and even in this case, so huge efficiency pains,

43:21 potentially worth way more is the uncovering process that you

43:25 have on those free emails. Oh, yeah. You would have just thrown

43:29 them away. Otherwise, how many times have I just deleted them

43:32 from this? I'm like, yeah, we're not finding these and they're

43:34 not going to Salesforce. So oops, no more. Not a problem

43:38 anymore. Then there might be who knows there could be a handful of

43:41 enterprise deals buried in those emails. So I think those are the

43:45 type of things where if you're not using AI, even let's assume

43:51 it took the same amount of time, even though it won't, you're

43:54 still leaving money on the table, missed opportunities,

43:58 miss level of depth to the analysis you're running. You

44:02 just have such an opportunity to take it to the next level.

44:05 Exactly. And I just I try to think I think I had a one of my

44:09 managers, Dan Walsh, when I was at BuzzFeed, he would always say

44:12 something would be really proud of something. And he'd be like,

44:14 Okay, yeah, what do you have to do to double it? I'd be like,

44:18 double it. I'm just so proud of what I just did. What do you

44:20 mean double it? Or he'd say the same thing and be like, well,

44:23 how about how are you gonna cut it in half now? And I'm like,

44:25 I'm just proud of where I got. But always having that mind of

44:27 like, well, what if what would have to happen for it to be like

44:32 kind of cooler or more intense, more excellent. And I think

44:37 having that in the back of my head being like, well, Claude,

44:40 you tell me what more can we do? What better research can we do?

44:43 That is the curiosity that keeps pushing you and to keep asking

44:46 Claude to get more out of it.

44:50The tech stack, and no MCP means no relationship

44:50 Curious, just because I'm a connoisseur of GTM tools, what

44:54 does the tech stack look like? And are there any tools that

44:57 you're super excited about right now?

44:59 Yeah, our tech stack is, let's see, we've got a lot of the

45:03 usual suspects. So we have Salesforce, HubSpot, Outreach.

45:08 We use Lucia for data enrichment, we use store leads,

45:11 store leads provides like e commerce data, we sell to e

45:13 commerce companies. Store leads, we use glyphic, they just

45:17 changed their name to Airspeed though. So Airspeed we use as our

45:20 call recorder. We of course use Slack, obviously the AI tools,

45:25 we use Spotdraft as our contracts management tool. We

45:29 don't have a CPQ system, I built and configured a some

45:33 lightweight something between Salesforce and Spotdraft that

45:35 serves as our CPQ solution. We use Stripe. What else do we

45:40 have tools wise? That's about it. We used to be on Webflow for

45:46 our website, but we cut that to build our website on Cloud Code.

45:49 So there are some like we've demoed a lot of tools, we've

45:52 done a lot of POCs with tools, we tried Artisan for like the

45:55 AISDR work, we were on zoom info for a while, we demo Cognizant

46:00 or to the POC with Cognizant. But right now we're running a

46:03 pretty lean tech stack, honestly, because we're we're

46:06 actually building again, most of our stuff. We're doing a POC or

46:10 demoing some deep line data right now, we use clay pretty

46:13 extensively. We've used saber in the past for some insights

46:19 tracking, I really like their stuff. But yeah, we're really

46:22 just buying data at this point, with the exception of like a

46:25 HubSpot and a Salesforce probably, and then trying to

46:28 make as many of our own products as we can. And that

46:32 makes sense.

46:33 No makes a ton of sense. I think now more than ever, like you

46:37 mentioned the CPQ, we've built custom CPQs for customers as

46:40 well. As if it's lightweight, there's a certain amount of

46:43 complexity where say, hey, no, let's go with a proper one. But

46:46 I do think it is a really interesting time for that build

46:50 versus buy discussion. Are there any tools that you feel like

46:53 are kind of on the chopping block or maybe categories of

46:57 tools? You don't have to name specific ones exactly.

47:01 I mean, any tool that doesn't have an MCP by now is not a tool

47:04 I'm looking to have much relationship with honestly.

47:08 Especially like, if I have to go into your tool to use it, I'm

47:11 not going to use it. And I'm not going to demo it probably.

47:14 And that's regardless of category. I don't care. I mean,

47:17 I think that goes for my personal life too. Like I run

47:19 my personal life out of terminal the same way I run my work life

47:23 out of terminal. So I mean, I installed like the WOOP MCP and

47:27 the Apple Music MCP recently, like I'm a freak, honestly. So

47:30 if you don't have an MCP, if you don't have a CLI, like I

47:33 don't see myself relying on you or seeing you as a really strong

47:37 partner. That's probably the first part. I am curious to see

47:43 what's going to happen with some of the conversational

47:44 intelligence. And I'm like, kind of split on what's going to

47:47 happen with it. So we love our transcripts. And we really use

47:52 our transcripts more than we use any of like the insights that

47:56 come from a lot of what what the platform will analyze for us.

48:00 We really reference like specific keywords or when

48:02 something specific was said at the point that a specific person

48:05 joined a call. There's a lot of specifics that we look for in

48:09 our transcripts. So I am trying to think if the transcript is

48:13 the most valuable part to me right now, I certainly am not

48:16 going to go build a conversational intelligence tool.

48:17 I'm not trying to compete with Gong that will not provide us a

48:19 competitive advantage. There's no point in doing that. But I

48:23 have re I have shifted my focus in terms of what do I actually

48:27 need from this whole system and like what's happening. I'm not

48:31 even totally convinced you need video and all of your call

48:33 recording. That's kind of a hot take. A lot of people like the

48:35 video. I don't think it's necessary. If you have the

48:38 transcript, if you're able to detect sentiment and tone based

48:41 off of the transcript or some other way. I think that's all

48:44 you need. I don't know what that means for the future of

48:46 conversational intelligence, but I know that it means I'm not

48:49 relying on it as much. It is more of just a mechanism to

48:53 capture the transcripts. I'm trying out Google meets a little

48:57 bit more right now. I'm trying to learn the Gemini

48:58 transcription service. Again, I'm not trying to rebuild

49:01 anything. I think it's really important that we focus on you

49:04 build your first MVP to prove that this is something that you

49:06 should go buy. But then you just buy we can buy we as the

49:12 companies have a little bit more leverage in terms of I don't

49:14 want to sign a full year contract. Maybe I'll sign a nine

49:16 month contract or six month contract. POCs are back so you

49:19 don't have to commit in the same way. That's how we're going

49:22 about it. We'll build our own MVP first. If it feels like it

49:24 will be that much more valuable, then we will go and buy

49:27 something. But we usually go through a few iterations to be

49:30 sure we know what we want before we end up buying anything.

49:34 Yeah, I'm with you on that. We use fireflies. Nice. Fun fact,

49:39 our COO, who's also my childhood best friend, but our COO was the

49:46 first hire at assembly AI. And they run the back end of so many

49:51 speech to text on if you know assembly then that's cool. So

49:57 sometimes we pick our products because he might still have some

50:01 level of investment there. But I think on the on the

50:06 conversational intelligence, I agree. I think I want to run it.

50:10 I want to run the intelligence my own way anyway. So as long as

50:14 I have a good transcript, and the only time you need video is

50:17 if you need to actually see something. But I think most if

50:20 you're demoing tools, even most of it has less to do with what

50:25 the UI is and or do with what the workflow and data

50:29 capabilities are. So totally do you need to see it?

50:34 I don't know. I'm not I'm not I'm not convinced yet. But times

50:38 before I'm usually wrong. So we'll see.

50:42Building a RevOps team now

50:42 Well, it's gonna change a ton as we go. But another thing that I

50:45 think I'm curious your take on building a rev ops team with

50:49 these new capabilities. What are the roles look like? I mean, I

50:53 used to have like an army of analysts, I used to have people

50:56 dedicated to adminning a tool. How are you thinking about

51:01 building your rev ops team with everything today?

51:05 I, the first thing I'll say, I mean, I, I have mostly worked at

51:09 early stage startups where headcount is not a privilege we

51:12 always have or budget that we can tap into. So when it comes to

51:17 building out a team, I will always start by saying people

51:20 follow pain. So whatever your pain is where you're spending

51:23 time where you can't recreate a workflow with AI, that's

51:27 probably where you need to hire first. Now when I think about

51:30 skill sets, and the profiles that are naturally going to

51:33 change because of the what our work is, and like honestly, what

51:36 our job is, is very different. Some of the fit, I mean, roles

51:40 wise, and I think the future of the rev ops, or one of the

51:44 futures I see happening is, there becomes a little bit more

51:47 of a bifurcation in terms of the strategy and the process and the

51:50 enablement, I think enablement is going to have a research too,

51:52 because we're spending all our time building stuff, creating

51:55 products, making sure people understand what is happening,

51:58 the information that's being fed to them, I think enablement is

52:00 going to have the next resurgence after rev ops, or

52:03 maybe we'll bring them along with us. But enablement is

52:06 going to be huge, I think that we're going to see a lot of

52:08 really strong enablement leaders that have been, maybe we're

52:11 early to enablement, like in the B2B in the tech space, the last

52:14 10 years, and now they're going to merge, especially the AI

52:16 pilled ones, I think are gonna be really, really powerful at

52:18 the orgs that they're at. But again, the process, the

52:21 strategy, even blurring, you know, with like forecast

52:24 management, that's still a little bit of a people skill,

52:27 working with the sellers working with the sales managers,

52:29 understanding where you're waiting your deals, I see all of

52:31 that as being like the traditional rev ops that stays

52:35 data analytics, systems, tools, automations, I suspect that

52:40 will move towards tech teams, data teams, I don't know if the

52:44 rev ops team or the go to market engineer is going to end up

52:46 reporting to the VP of data, I don't know. But I see similar

52:50 patterns in the type of work that's happening. Obviously,

52:53 here at Vermont, like I manage most of the tech stack, it

52:56 wouldn't be that much of a leap for me to add like the other

52:58 five tools here. So you can see how someone who's been in go to

53:01 market engineering, you know, if you are a data analyst, I said

53:04 this actually last week to my team, like we are, we're

53:07 basically a data integrity team that's really good at

53:09 automations and tools, right? Because everything that the

53:12 business runs off of, it's our job to give the business what it

53:15 needs to make decisions today. Almost all of that comes from a

53:18 system. And those systems are built on the processes and the

53:22 strategies and the rules that we have set. That is so much work

53:25 that I could see it staying with someone else, especially as some

53:28 of these like AI companies are growing exponentially. And then

53:31 you really have a more technical team could be go to market

53:34 engineers could be someone like me, I still consider myself

53:37 non technical, but like that sounds very interesting. Working

53:40 with the data and analytics team being closer, you know,

53:43 especially with all these usage based businesses and consumption

53:46 pricing, like the data warehouse is also a go to market tool, it

53:51 captures metering and it captures rate cards and usage, it

53:53 calculates billing, it dictates the rev rec schedule, it has a

53:57 lot. So it's not a big leap to say, actually, the people who

54:00 have been managing 60 to 80% of the tech stack, and who have

54:04 also been responsible in communicating how to use it,

54:07 maybe we should also give them the other technical tools that

54:10 we also own, because all of that reporting gets fed in

54:13 together. So that's where I see like this AI pill technical

54:17 aspect of rev ops growing, and the softer more people elements,

54:23 and really the first line of defense for rev ops before we

54:25 know what to build, I see that staying as a little bit of a

54:27 separate thing.

54:29 You know, what I've seen as a, as a trend recently that I think

54:33 would align with your hypothesis is seen a few heads of rev ops

54:37 move into a COO position. And in that COO position, it would be

54:43 very natural to say, Okay, well, I have a VP of rev ops, maybe

54:47 maybe product ops or engineering ops and then but within one

54:51 executive owning the entire tech stack infrastructure of the

54:54 company.

54:56 Totally. That's I think that's going to happen. It's not that

54:59 many more tools. And when you think of like, who has the most

55:02 experience, again, in communicating how to use a tool,

55:04 how to uncover it, but also what tool we should use how to manage

55:08 it, the privacy concerns within a tool, I could see it really

55:11 following falling, and being, you know, consolidated into one

55:15 team for sure.

55:16 And the go to market data is the messiest data. It's the most

55:19 difficult data to work with. I mean, if you look at product and

55:22 engineering data, or financial data or agent, that data is so

55:26 clean, and so easy to work with. So we've already done things on

55:30 hard mode, we might as well just go get the easy stuff too.

55:33 Yeah, seriously, that's crazy to think that like the ERP would be

55:36 easier than I believe it. Yeah, I think when I know building out

55:42 the team to and like some characteristics, one thing that

55:45 I've been thinking a lot more about recently is you especially

55:49 like we're in these fast growing orgs, we're in the age of AI, a

55:52 lot of teams are relatively flat, like our or get Vermont, I

55:56 have a team, but we're essentially a flat org for the

55:58 most part, which is really exciting. And it means you can

56:01 be a manager, but you're also expected to be an IC, which is

56:03 great, because I love to build and I love to be doing my work.

56:07 But the, the importance of building out loud, I think is

56:11 even more important, because no one's going to come in and check

56:14 on you, everyone is running a million things on their own, you

56:17 need to be constantly updating and communicating. This is what

56:20 I'm spotting. This is what I'm building. This is my approach.

56:23 Even just last week, I shared something we've a shared AI

56:25 enablement channel where we all just drop in like, this is what

56:27 I've learned, or this is what I'm doing. And I dropped in, honestly,

56:31 I was talking to myself more than I was talking to anybody

56:33 else. But I was like, this is how I decide when to use code

56:35 versus codecs. And in the middle, I'm like, I don't know if

56:37 this is right. But this is how I'm choosing it. And this is

56:40 what's making me say that. And then we'll just end up having a

56:42 whole conversation from that. So being able to say, this is what

56:46 I'm doing, I don't know if anybody's listening, I don't know

56:48 if this matters to anybody else. But it was helpful for me. I'm

56:51 gonna at least try to share it. I think that's really, really

56:54 important for the next several years in particular.

56:57 Well, in the spirit of sharing, I know you prepared a couple

57:00 demos, I'd love to see a few things in action. Maybe we do

57:05 need video for this one, maybe the transcript won't just

57:07 suffice.

57:09Infrastructure skills and the 50% context rule

57:09 Fair enough. Alright, so I've got a few different skills that I

57:14 thought I would share. I'm going to start with some skills. So

57:18 again, I operate mostly in terminal, I manage my email from

57:21 terminal, most of my slack, anything that comes to me, I try

57:25 to start from terminal, even if I take it somewhere else. So I

57:28 actually don't work in the terminal app, I work in CMUX.

57:31 It's really nice, because you can have like lots of different

57:33 quadrants. And then I have different workspaces along the

57:35 side. So what you're seeing here, it is terminal, it's just

57:38 in a different UI. So I'll come in here and I'll open up Claude,

57:41 Claude code. And the skills that I'm gonna walk you through

57:44 today, because I think these are the ones that are easiest for

57:46 many people to grasp, because it's like a shared process, I'm

57:49 going to go through some skills that fall into what I consider

57:52 the infrastructure bucket. So to give you a little context on how

57:55 we got here, this is me and probably April or May. I'm

58:00 realizing, okay, nothing out of my folder structure is correct.

58:03 None of my file structure is correct. Every time I think

58:06 Claude remembers not to use an m dash, it sends me an m dash in

58:09 the next next message. I was having all this frustration. And

58:12 I went to call and I said, What do you need from me in order for

58:15 you to work exactly or near exactly as I expect you to? And

58:19 we went through all these things tone and voice, folder

58:21 structure, what to push to GitHub, what's for, you know,

58:24 personal work or personal folder that might be on my desktop,

58:26 whatever. And from that, that's where we built all of these

58:30 infrastructure skills. So a few of them before I even start

58:33 getting into it, I want to pull up here and show you right down

58:37 here, you'll see I added this usage bar, or I call it a status

58:40 bar at the bottom. So I met with a friend of mine who's a go to

58:44 market engineer. And he told me something that changed the way I

58:48 think about how to run my sessions. He said, you should

58:50 never run a session past 50% context, because that's when

58:53 Claude and even codex really starts to degrade. So I said,

58:57 Okay, great. How do you know when you're at 50% context? And

59:00 he said, Oh, you can just ask Claude, I was like, I'm not gonna

59:02 ask Claude. Instead, I'm gonna create a breadcrumb for myself,

59:05 where I'm going to add a marker here. So I'm going to say, let

59:08 me just say, Hey, Claude, and let's see if we can get it

59:11 running, because then there's a little status bar that will load

59:14 here, and it'll show you how much usage. Here we go. So just

59:18 pulling this up, just x out of that is showing me that I've

59:22 already used 8% usage for what's in this chat in this session. So

59:26 now I know when I get to 51%, I got a clear context and I got to

59:29 start a new session. So that became my next friction point,

59:32 right? So first, it was the clearing context at the right

59:35 usage amount. So I created this little breadcrumb for myself, a

59:38 visual guide to know when to switch. Now I want to make

59:41 clearing context as quick as easy as possible. So in order

59:44 to do that, I'm going to pull something up this we might want

59:47 to demo output podcast prep. I'll explain this skill after. But

59:53 what I'm going to pull up here is we're going to pretend that

59:55 I'm in the middle of a session where I am preparing for this

59:59 podcast interview. Because of course, when Anthony reached out

1:00:01 to me, I think, send me an email, probably send me an email

1:00:05 or send me a LinkedIn DM, that hit my work email. So I go into

1:00:08 Claude here and I say, Hey, I just saw an email come in from a

1:00:11 LinkedIn DM from someone talking about a podcast sounds really

1:00:13 interesting. Can you take a look at it? I can't remember if Claude

1:00:16 can actually read the email or the inbox from the email, but

1:00:21 that's the idea. And that's like how I try to work. So here you

1:00:24 can see, let's pretend I'm right in the middle of this session,

1:00:26 right here, where I'm like, Okay, Anthony, and I are going to

1:00:29 talk about robots greatest hits, let me scrape, you know, our

1:00:32 projects dash or projects database, make sure I'm prepared

1:00:35 with everything that we need. Let's see. Anthony's share

1:00:38 Google Doc with me, yada, yada, yada. Okay, great. I created this

1:00:41 doc, I feel really prepped. I'm at 10% context, let's just

1:00:44 pretend that I'm at 51% context now. And before I prep, you know,

1:00:48 my talk track, I want to clear context. So in order to do that,

1:00:52 make that really easy, because I'm not a senior software

1:00:55 engineer, I don't know how to make sure that Claude always

1:00:58 documents every decision we made, or writes a really clean hand

1:01:00 off doc, so that the next time I come in, and I want to work on

1:01:03 prepping for this podcast, I can just pick up right where I left

1:01:06 off, and I don't even need to lose a beat. So that's what I was

1:01:08 trying to solve for, right, because I constantly need to

1:01:11 clear context. So the first thing I did was I created this

1:01:14Wrap and pickup: 250 times a month

1:01:14 skill, it's called wrap, you can see here, it's called end of

1:01:16 session wrap summarizes what was decided and built updates, a hot

1:01:19 cache and creates a compact handoff doc. So every single

1:01:23 time I am in a session, and this goes for my personal life and my

1:01:26 work life, I use CMUX for both. I'm done with a topic or maybe

1:01:31 Anthony, I'm prepping, and I'm gonna go for lunch, and I'm gonna

1:01:33 close this session, I'll come here and I'll say wrap. Right?

1:01:37 And I think it might respond to me and say there's nothing to

1:01:39 wrap here, because technically, we didn't make any changes. But

1:01:43 actually, hold on. Let's do yeah, let's do Okay, update, the

1:01:48 podcast got pushed to Friday, let's just make a change. So

1:01:53 before I actually wrap the session, now I'm going to make

1:01:56 this change so that it has something to wrap. But in this

1:01:58 wrap skill, it's going to update, you can see this, see this

1:02:01 handoff, Anthony podcast, MD, this is the file that created the

1:02:05 first time I was working on this podcast prepped. So it created a

1:02:09 file. And that's how when I come in, I'm like, hey, let's pick up

1:02:12 that project. I will just say, hey, pick up Anthony podcast.

1:02:18 Okay, update. Sorry, it's not good. I'm in plan mode right now,

1:02:24 because I'm trying to think of ways to maintain my spend a

1:02:27 little bit better. So I default to plan mode all the time. But

1:02:31 here, it's going to update a dock. These are some of my hooks

1:02:35 that are in place to stop bad things from happening. But here

1:02:40 we go handoff files are owned by wrap, let me invoke it

1:02:42 properly. So now because I just said, okay, an update to this

1:02:45 project is that the podcast is happening on Friday, it's

1:02:48 updating the dock. Now as it wraps this whole session, it's

1:02:51 going to go through and run through a few other skills that

1:02:54 I have. Like for example, I have a PM plugin that updates all of

1:02:58 our projects in notion for the RevOps team creates new

1:03:01 projects, creates problem statements, gathers context on a

1:03:05 project before we launch it. And then here we go. This is exactly

1:03:08 what it looks like. So session wrapped what we did picked up

1:03:11 the Anthony podcast recording got pushed to Friday. And then

1:03:15 it shows me here, these are the files that we actually wrote

1:03:17 documents every decision we made here, anything that I said, Oh,

1:03:20 we'll have to do that in the future. And now it's time for me

1:03:23 to clear context. But before I hit clear, I need to know

1:03:27 exactly how to start this conversation again. So I'm going

1:03:29 to copy, I'm just going to copy this, go clear, it's going to

1:03:33 clear, we're back to zero, and I'm going to paste and it picks

1:03:37 up Anthony podcast exactly where we were 10 seconds ago, and the

1:03:41 whole thing's gonna load, it's gonna say picking up the session,

1:03:43 this is where you left off, it was moved to Friday, and it'll

1:03:46 take you know, I don't know, probably five or 6% of our

1:03:49 usage. And that's how I pick up immediately between sessions. So

1:03:53 I'm always staying on the strongest version of Claude, I'm

1:03:56 not degrading over time as I get into like 70 or 80%. And then

1:03:59 I'm practicing this amazing, I guess, engineering hygiene by

1:04:05 documenting everything in exactly the right way, so that I

1:04:07 can move really quickly. This is one of my favorites. I when I do

1:04:12 like look at my skill count, I have used wrap and pickup, I use

1:04:16 them about 250 times a month, which is pretty crazy.

1:04:20 Well, and also not just the technical aspect of running

1:04:24 these, but I think there's still a human element to running your

1:04:27 workflows and agents and running your sessions. And there's

1:04:31 something calming about saying, Okay, I am done with this. And

1:04:37 also, like I know that everything that I was working on,

1:04:41 I can go pick up again, where if you're just running like hundreds

1:04:45 of random sessions, you're looking to go back in, if you

1:04:48 start a new one, you're gonna kind of like re prompt the

1:04:51 entire process all over again. So it's just good PM practice as

1:04:57 well.

1:04:57 Exactly. And to go back to the like three, you say it three times,

1:05:02 when I first came up with the wrap skill, I think I cleared

1:05:07 it. But when I first came up with the wrap skill that came out

1:05:10 of me saying to Claude, I don't want to ask you to do the same

1:05:14 things every time. If you need to update the readme and you need

1:05:16 to update the memory doc, I need you to know to do that by

1:05:19 yourself. I'm not going to sit here and remind you I need one

1:05:22 word to invoke it. And that's when Claude was like, wait,

1:05:24 that's what a skill is. So then we turn this into a skill, I have

1:05:27 another skill, I won't go through all of them. But I'll

1:05:28 show you. I have oh, it's the plugin is called the skill

1:05:32 factory. But I have the skill factory has multiple skills

1:05:36 within it. And the skill factory is all about like when I spot

1:05:39 bugs in my skills, I capture a bug. So it logs a bug. And then

1:05:43 I'll go q fixes. That's where it is. Hold on. Change them. Okay,

1:05:52 so apply fixes. This is the skill that reads through all of

1:05:56 the logs of the bugs and other skills, and then goes through

1:05:59 and checks how to fix them. So this is like an infrastructure

1:06:01 skill where they're meant to make sure that I'm getting the

1:06:05The core tools tracker for silent failures

1:06:05 most out of Claude. Another one, this core tools down here, I

1:06:08 have some, you know, like a chief of staff that runs every 20

1:06:10 minutes or so. And I keep kept getting from a chief of staff,

1:06:13 like you don't have any new emails, you're good. And then I

1:06:15 realized four hours later, the Gmail MCP had broken or, you

1:06:19 know, wasn't connected. So in order for that to not happen, and

1:06:22 for me to not be, you know, misled by Claude, I created this

1:06:26 core tools tracker. So if any one of my five tools, like email,

1:06:30 calendar notion slack granola are offline, I know about it. I

1:06:35 also have a recurring MCP heartbeat to check to make sure

1:06:38 that I have what I need. We've created a lot of these skills

1:06:41 that really just like work along beside you. And again, why I

1:06:45 like sharing these is because no matter what you're doing in

1:06:48 terminal, these skills will be valuable for you. And it's

1:06:50 something that you can start building on.

1:06:53 I'm a big fan of that core tools one that you mentioned,

1:06:56 because there are so many times where I'm running something and

1:07:00 yeah, one of my tokens expired or something. And now it's not

1:07:04 connected to that. And it's trying to run it without it or

1:07:08 so, especially if you have any routines or ongoing workflows

1:07:13 that are scheduled. There's so many times I have routine set up

1:07:17 that just don't work, right? And then I have to go in and manually

1:07:21 do it. So it's a huge vein. I'm stealing this one. I hope that's

1:07:24 okay. All yours. Let me show you a few other ones. This one I

1:07:27 added recently, this only works in Claude codex doesn't expose

1:07:32 it. But this is how much I've spent in this session. And then

1:07:35 this right here is my rolling 30 day average. I'm not really

1:07:38 sure how accurate it is. So don't quote me. But then I don't

1:07:42 know if I want this one. I don't know if this is a good one for

1:07:45 me. It's helpful for me. And I use these to, you know, spend

1:07:51 ones in conjunction with the model. And it's helpful to just

1:07:54 see like what spikes bright price, like when I scan through

1:07:58 notion that spikes price a lot, but when I write to notion, it

1:08:01 doesn't take up as much. But when I scan through slack, it

1:08:03 doesn't take up as much usage. So by adding in the model down

1:08:07 here, I, again, it's still on me to do the changing of the model,

1:08:11 but I have these visual cues and visual reminders for me to not

1:08:15 forget things that I need to get better at or gaps that I have.

1:08:18 Like I haven't spent or more recently, I've been spending

1:08:21 time trying to diversify my models. But this is this was the

1:08:24 first step I was like, let me at least understand what I'm on

1:08:27 when I'm on it. That was part one. We've got a few other ones

1:08:32The RevOps operating system and releases

1:08:32 that I'll just show you we created this ruse rev ops

1:08:36 operating system. This is a plugin with a series of skills

1:08:40 that are all internal team management for our team. So

1:08:43 before we start a project, we will always write a problem

1:08:47 statement similar to like a product team, you start with a

1:08:49 user statement, a user journey, what's the impact, we'll do the

1:08:52 same thing, you know, you want me to import a lead list, I'm

1:08:56 going to create a problem statement first. And then from

1:08:58 there, we'll generate a project. From there, then we will scope

1:09:02 the project, then we will actually make the tasks that are

1:09:04 related to the project, and then we will actually execute on it.

1:09:07 So we went through, let me see if I can go through a few of

1:09:10 these. For example, this new project skill. So this new

1:09:15 project creates the actual problem statement in notion that

1:09:20 lives in one database, that page is related to our projects

1:09:23 database, this skill creates projects for us and it follows

1:09:27 the same instructions, scan slack for context scans granola

1:09:30 scans all of our tools to add to the problem statement that we

1:09:34 surfaced, then it generates the problem, the project itself in

1:09:39 notion for me, like right in the format that we need, that it

1:09:43 generates the tasks, it does everything for us. And we'll do

1:09:46 I think I've got another one in here about sprints. But we

1:09:50 operate on the sprint schedule. So you can see here, when I will

1:09:55 post that we have a new RevOps sprint starting and like, what

1:09:58 are we working on this week, I'll start by drafting with the

1:10:01 RevOps sprint launch. And we'll have one for the wrap. We also

1:10:05 have this other one called RevOps releases where again, we

1:10:08 try to operate as much of a as a product team as possible. So

1:10:11 whenever you know product ships a new feature, they will post

1:10:14 like ship shipped in the ship channel. We had FOMO we want to

1:10:18 do the same thing. So now in our RevOps slack channel, once we

1:10:21 release a project and you know, there's a difference for our

1:10:24 stakeholders, we'll do a RevOps release, we have some pictures,

1:10:27 it creates you know, a short little memo that goes in notion

1:10:30 it's linked to the project, it's linked to the documentation.

1:10:34 Anybody who wants to learn about a project that we released is

1:10:36 going to have all these different artifacts to work off

1:10:38 of and we approved all of the language that goes in them.

1:10:41 That's what we do from terminal from these skills, but we did

1:10:43 not manually write them. And we certainly did not have the time

1:10:46 to go get all the context to build it out and get the right

1:10:48 project.

1:10:52 These RevOps ones are good. This is also where we have, let me

1:10:54 show you our pipeline ops. So that's what I was talking about

1:10:57 earlier. New pipeline review them, add a contact to a

1:11:02 specific campaign, review your pipeline before you go into a

1:11:05 pipeline meeting. So we've got all these different, you know,

1:11:09 habitual routines that our team is doing. And we're trying to

1:11:12 really perfect the process and the framework for that skill for

1:11:15 us before we start building out like even more agents that you

1:11:19 know, are external facing as well. This is a way that we're

1:11:21 using to manage our blast radius with a lot of these AI

1:11:25 capabilities is we're really focusing on what is the internal

1:11:28 stuff that we feel confident about and then we will expand to

1:11:30 more things.

1:11:33 Smads, this is fantastic. I really appreciate you walking

1:11:36 through these and sharing them. And I think the biggest takeaway

1:11:41 is how you're approaching the infrastructure component of your

1:11:45 agents and workflows and making sure the MCP is connected, are

1:11:49 we consistent with how we're, how we're launching these agents

1:11:53 or building these capabilities. And I think that's where the

1:11:56 teams that are making the investment in that component of

1:12:00 AI are going to start to really, really run fast and make a big

1:12:05 gap between the ones that are not.

1:12:07Treat agentic infrastructure like Salesforce

1:12:07 I agree. And I think I said this a little bit earlier too, but we

1:12:10 need to think of these platforms and these tools and what they

1:12:14 can do for us in the same way as bringing on a Salesforce. This

1:12:17 is now a cornerstone of your tech stack. And we have

1:12:20 maintenance days, we have code freeze periods for Salesforce,

1:12:23 we have blackout dates, all of that is probably going to end up

1:12:27 also applying to the company wide agentic infrastructure, as

1:12:30 well as your own personal operating system. And if I want

1:12:33 to make sure that my team is pulling the right context for a

1:12:36 project or writing the right type of problem statement, all I

1:12:39 got to do is write a skill. And then we use it.

1:12:45 Amazing. Smads, thank you so much for being on the podcast

1:12:50 today and sharing everything. I love the story of seeing that

1:12:54 waterfall chart, getting the spider bite and finding your way

1:12:57 into ops and then all of the things that you're doing on the AI

1:13:00 front really, really pushing the limits and sharing a standard of

1:13:03 what good looks like for RevOps. So thank you for being here. And

1:13:08 I can't wait to see what you build next and hope that you can

1:13:12 share some of these with audience as well.

1:13:14 Definitely, definitely. I will include a link to my GitHub where

1:13:17 I have, whether it's the actual skill or the prompts that you

1:13:19 can just download and use and then you have to share it with

1:13:22 someone else. That's the rule. You can't just keep it to

1:13:24 yourself. We got to share.

1:13:27 Too generous, too generous. Thank you again.

1:13:29 No, thank you for having me. This has been awesome. I will

1:13:31 geek out anytime. So thank you. Thank you for having me.