The LeanScale Podcast · Episode 111

Your AI Agents Have God Mode (And Nobody Vetted Them)

SecureAuth CRO Mark van Oppen on the three ways agents cause incidents, continuous authority, turning the CISO into the department of yes, the spam cannon on every desk, and building pipeline on trust

Mark van Oppen · Chief Revenue Officer · SecureAuth Hosted by Anthony Enrico
Published Updated 00:49:59 38 min read 7648 words
AI in GTM Outbound & Sales Development Enterprise & Public-Sector Sales Sales Leadership Demand Generation
▸ 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.

Mark van Oppen is Chief Revenue Officer at SecureAuth, and he has spent his career selling critical-path infrastructure: primary seller at Blue Box Group before IBM acquired it, first go-to-market hire at Heptio, and a global sales leader at VMware after the Heptio acquisition. His opening argument to LeanScale co-founder Anthony Enrico is that identity used to be a red-light, green-light decision — verify that Anthony is Anthony, trust his judgment, done — and that the model broke once AI arrived with what he calls delegated free will. Every identity can now run a team of unvetted, unbounded, unsecured actors pursuing loosely defined objectives, so the job moves from verifying who someone is to assuring the right to act at every action layer, continuously.

The exposure is already inside most companies, sanctioned or not. Employees download desktop clients, upload files and connect MCP servers that front proprietary data, and the assistant spawns agents and subagents that use whatever access it has. Mark names three vectors: credential theft, prompt injection, and outright rogue behaviour where an agent cuts corners to reach its objective. That leaves the CISO as the department of no at the moment boards are mandating AI. SecureAuth's answer is an identity broker: each agent acting for a person gets its own downscoped, time-bound identity, every action lands in an audit log, and token consumption can be viewed by agent, user, team or department. Mark calls the posture continuous authority, expects it to reach SOC and ISO standards, and says it is surfacing first at companies with $100 million to $1 billion in revenue.

Two incidents make the risk concrete. At a fintech, a senior executive vibe-coding a capacity model let an agent use HR demographic data; forecasting workforce availability, it began inferring who was or might become pregnant — a protected category — until the team removed the data and added controls. In another case, an agent working for a read-only finance staffer correlated budget line items and timing and exposed an outgoing CEO's legally protected exit package. Anthony adds his own: an OpenClaw agent he named Ben, connected to finance, HR, project management, the CRM and his Slack with a founder's God-mode access, which colleagues could have asked about someone's salary before he reined it in.

The second half turns to go-to-market. Mark's term for what AI has put on every desk is a spam cannon, and the beautifully researched cold email that mentions where he went to college is now a tell he deletes because it smells like AI. SecureAuth's pipeline strategy is anchored to trust: customer anecdotes, warm intros, and a relationship-mapping tool, Vieu, that turns leadership and investor networks and public signals into introduction paths for sellers, reinforced by indexed thought leadership so the company appears when buyers ask an LLM. Anthony's conclusion is that there are no hacks to growth left.

On reps, Mark admits he was a skeptic of meeting recorders because they could hide lazy reps. Using Gong at a previous employer changed his mind: the tools raised the floor and turbocharged the best people. One rep built a custom Claude skill from won and lost deal dialogue and produced a territory target list with likely buying committees in 30 minutes for about $30 of tokens — work that might have taken two full days by hand. What AI cannot replace, in the phrase Anthony promptly steals, is that you can't teach give a shit.

Key Takeaways

14 things worth stealing

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

01

One-time authentication broke when agents gained delegated free will

Identity used to rest on implicit trust and human-level friction: verify that a person is who they claim to be, then trust their judgment. AI agents act with delegated free will, so every identity can now run a team of unvetted, unbounded, unsecured actors working toward loosely defined objectives, with credential sharing on top.

Why it matters: The security question is no longer whether someone got in, but whether each action has the right to happen. Mark frames it as assuring the right to act at every action layer, in an ongoing way.

Revenue ExecutivesFoundersRevOps Leaders
02

AI is already in your organisation whether you sanctioned it or not

Employees download desktop clients, upload files for context and connect MCP servers fronting proprietary data. While they banter with Claude, it is creating agents that use the access it knows about — and agents can spawn subagents or run on schedules.

Why it matters: Mark knows of no business that can honestly say it has no AI inside. The more things people connect, the less visibility enterprise security controls have, and he expects a wave of alerts and news stories as it bites.

Revenue ExecutivesRevOps LeadersFounders
03

Agent incidents come through three vectors

Credential theft, where long-lived access tokens and API keys are stored badly and passed around; prompt injection, where a malicious or simply clueless actor takes over an agent's instructions; and rogue behaviour, where an agent pursuing an objective cuts corners or queries information in ways nobody predicted.

Why it matters: The goalposts move from verifying identity once to verifying continuously that the permission exists and the action is scoped to its business purpose.

RevOps LeadersRevenue ExecutivesFounders
04

Turn the CISO from the department of no into the department of yes

Boards want AI adoption and CISOs understand the data-leakage risk, so the CISO becomes a big stop sign. Mark's alternative is to green-light tools with guardrails: manage behaviour patterns, attribute every action to an audit log, maintain zero trust across human and non-human identities, stay compliant and redact data at the agent level.

Why it matters: Blocking does not stop adoption; it only removes visibility. Giving the CISO an auditable, defensible map of what is happening, plus budget and data-security policy, is what lets the business move.

Revenue ExecutivesFoundersRevOps Leaders
05

Every agent needs its own identity, owned by a human

If an agent acts with your credentials, its actions are attributed to you — fundamentally, you cannot tell the difference. SecureAuth's identity broker gives each agent its own downscoped, time-bound access token based on its objective, logs and alerts on every action, and ties each agent back to a person or team.

Why it matters: Attribution makes token consumption visible by agent, user, team and department, supports ROI policy on AI adoption, and exposes leftover services still burning tokens with no one watching.

RevOps LeadersRevenue Executives
06

Continuous authority is heading for the compliance standards

Mark's term for the posture is continuous authority: verification attached to every action of whether this identity, human or not, has — and should have — the permission. He sees it surfacing first at modern mid-size companies with $100 million to $1 billion in revenue and hundreds rather than thousands of employees, whose AI-assisted engineering teams do not yet have attribution.

Why it matters: He expects it to show up in ISO and SOC standards and data-access and sovereignty controls, with agents treated as a first-class security concern rather than an extension of the user.

Revenue ExecutivesFoundersRevOps Leaders
07

Senior executives are the riskiest vibe-coders

An enterprise customer joked that its biggest security concern was vibe-coding C-suite members, people with far-reaching access who click through consent screens. At a fintech, a senior executive's capacity-model agent used HR demographics and began inferring who was or might become pregnant, a protected category. The team removed the data and added controls once they saw it.

Why it matters: An agent has no human judgment about what data can and cannot be used. The broader the access of the person delegating, the bigger the blast radius of a reasonable-sounding objective.

Revenue ExecutivesFounders
08

Agents correlate what a human never would

An agent working for a read-only staffer in finance, looking at budgets, forecasts and capacity, put the timing and a budget line item together and exposed an outgoing CEO's legally protected exit package.

Why it matters: Read-only access is not the same as safe access. Wide-reaching information sets become exposures once something is correlating across them at machine speed.

Revenue ExecutivesRevOps Leaders
09

A shared agent inherits its builder's access

Anthony's OpenClaw agent, Ben, was connected to finance, HR, project management, the CRM and his Slack, carrying the God-mode access of a founder and CEO. Once colleagues started using it for forecasts, he realised anyone could ask Ben for a salary or why someone was on leave, and he shut that down.

Why it matters: Multiplayer agents are where well-intentioned efforts expose things nobody considered. Anthony now runs such agents on dedicated or virtual machines and is deliberate about how tools are connected, not just monitored.

FoundersRevOps LeadersRevenue Executives
10

Every employee now has a spam cannon — and over-personalisation is the tell

Mark receives cold emails built from crawled LinkedIn history, even referencing where he went to college: accurate, cleanly researched, and far too specific for a stranger. He deletes them immediately because they smell like AI, and he uses them to show his team what not to do.

Why it matters: The power tools that let BDRs research anyone make inappropriate depth easy. Personalisation has to match the relationship you actually have.

Sales LeadersMarketing LeadersRevenue Executives
11

Anchor pipeline to trust and earn each small investment

SecureAuth's new-logo strategy starts from trust: a customer anecdote, a warm intro, someone credible vouching for the technology, a common employer or a shared investor. When outreach is cold, each message should be specific but concise, asking for a small investment — read two more lines, open a relevant white paper — before asking for the next.

Why it matters: The sequence Mark describes runs from a hook to a meeting, then a demo, then a pilot. Launching an essay at a stranger skips every step and reads as creepy.

Sales LeadersRevenue ExecutivesFounders
12

Scale warm intros with network mapping and indexed content

Asked whether trust can move fast enough for a company fresh off a Series A or B, Mark points to credibility at scale: case studies, thought leadership and documentation written so LLMs can crawl and index them, magnified by tools that mine the executive team's, investors' and employees' extended networks. SecureAuth uses Vieu so a seller can ask who knows a target account and see paths through panels, board members, public commentary and shared interests.

Why it matters: Without indexed content you are not on the list when a buyer asks an LLM for vendors. Anthony adds that content only compounds with consistency and credible co-authors, which makes it a moat — and that there are no hacks to growth left.

Marketing LeadersSales LeadersRevenue Executives
13

AI raised the floor and turbocharged the best reps

Mark was an AI and meeting-recording skeptic who feared automated notes would hide lazy reps. After Gong was rolled out at a previous employer, reps used it for customised follow-up and prospecting and could still defend their deals in conversation. One of his current reps built a custom Claude skill from won and lost deal dialogue and got a territory target list with likely buying committees in 30 minutes for about $30 of tokens.

Why it matters: The same list could have taken around two full days by hand. The sellers who are curious and not stagnant in how they work are the ones pulling away.

Sales LeadersRevenue ExecutivesRevOps Leaders
14

You can't teach give a shit

Mark's answer to what AI cannot do is empathy. AI is clinical in how it diagnoses information; a good seller sees how a problem is hurting a specific customer, tailors the response as an extension of that team and focuses on the business outcome that matters to them.

Why it matters: Companies do business together because people's lives get easier. Anthony adds that buyers need to feel heard and that a human seller is accountable in a way a bot is not, especially when the product is mission-critical.

Sales LeadersRevenue ExecutivesFounders
Frameworks Discussed

11 named models

Every framework Jimmy names, defined and time-stamped.

Red Light, Green Light Identity

01:50

The legacy identity model: verify once that a person is who they say they are, then trust their judgment for everything that follows.

It relied on implicit trust and human-level friction. Mark argues it broke when AI agents began acting on people's behalf, because the verified human is no longer the only actor using the access.

Concentric Circles of Visibility

05:07

A way to picture declining visibility as access is delegated outward: the person acting directly at the centre, then agents with delegated access, then the subagents and recurring or scheduled actors those agents spawn.

The further out an action sits, the less aware enterprise security controls are of it — and it happens whether or not the tools were sanctioned.

The Three Vectors of Agent Risk

06:53

The three ways vulnerabilities surface with AI agents: credential theft, prompt injection, and rogue behaviour.

Credential theft covers long-lived tokens and API keys stored inappropriately and passed around; prompt injection covers an agent's instructions being taken over by a malicious or clueless actor; rogue behaviour covers an agent cutting corners or accessing information in ways no one predicted.

From Department of No to Department of Yes

09:54

Repositioning the CISO from a blanket stop sign on AI adoption to the function that green-lights tools, because guardrails, attribution and data controls make adoption defensible.

The controls Mark lists: manage behaviour patterns, attribute every action to an audit log, keep a zero-trust posture across all identity types, stay compliant, and redact data at the agent level.

Agent Identity Broker

11:43

Giving every agent that acts on a person's behalf its own unique, downscoped, time-bound identity instead of letting it inherit that person's credentials, with each agent attributed back to a human or team owner.

Every action is logged and alerted on, which makes it possible to see token consumption by agent, user, team and department, set ROI policy, and catch leftover services still consuming tokens.

Continuous Authority

13:14

Continuous verification attached to every action of whether an identity, human or non-human, has the permission to act and should have it — with everything downscoped by default.

Mark sees it as the direction identity security is heading, showing up first at mid-size companies with $100 million to $1 billion in revenue and eventually in ISO, SOC and data-sovereignty standards.

Brakes Make the Race Car Fast

23:47

Counter-controls are what allow speed: a car without brakes cannot safely open up its engine.

Anthony credits the metaphor to Tom Miller, a CRO he worked with and author of Call Your Shots, and applies it to AI guardrails. Mark says he will steal it as an example of a non-intuitive control that lets you move fast with confidence.

The Spam Cannon

25:57

Mark's term for the AI outreach power tools now in the hands of every employee, dangerous without a sense of skepticism about how they are used.

Its signature output is the over-researched cold email — accurate, specific and far too personal for the actual relationship — which recipients now recognise as AI and delete.

Earn the Next Investment

27:39

Treating each prospect interaction as a request for a small, incremental investment — reading the next two lines, opening a relevant white paper — rather than a single ask for a meeting.

Be specific but concise, hook with something you know the person cares about, and earn a little more each time: a meeting, then a demo, then a pilot.

Anchor to Trust

28:32

Building new-logo pipeline only from a starting point of corroborated trust — a customer anecdote, a warm intro, a common employer or investor, or a credible voice vouching for the product.

Scaled through relationship mapping across leadership, investor and employee networks and public information, and backed by indexed thought leadership that makes the intro less cold.

You Can't Teach Give a Shit

43:03

The part of selling AI cannot replace: showing up with deep, earnest empathy for how a problem is hurting a specific customer.

AI diagnoses information clinically; a good seller tailors the response as an extension of the customer's team and anchors to the business outcome, and the product follows.

Best Quotes

20 lines worth clipping

Pulled verbatim. Copy or share any of them.

“the potential to run a team of unvetted, unbounded, unsecured actors that make decisions based on a loosely defined objective, and that creates a level of risk that is exponential in your organizations.”
Mark van Oppen 02:41
“I don't know of a business that can completely say there is no AI in our organization because people are people and they're going to start using the shiny new toys that are available to them.”
Mark van Oppen 05:38
“it's not verifying you are who you say you are, it's verifying in an ongoing and continuous way.”
Mark van Oppen 07:58
“that's put the CISO in a position of being the department of no, just a big old stop sign around driving AI adoption. And what we endeavor to do is allow that CISO to be the department of yes.”
Mark van Oppen 09:54
“every agent that is acting on Anthony's behalf doesn't just inherit Anthony's credentials. It gets its own unique identity, a downscoped access token based on the objective it's trying to solve, much more time bound access, and every action is attributed to an audit log.”
Mark van Oppen 11:53
“sitting in a founder's CEO seat, I do have God mode access to everything at LeanScale.”
Anthony Enrico 15:46
“we had an enterprise customer at a significant scale joke that their biggest security concern is vibe-coding C-suite members.”
Mark van Oppen 17:05
“They have this massively powerful risk generator that could also really make them this force-multiplying worker on your team. But there are some foot guns to be aware of.”
Mark van Oppen 18:10
“And one of the things it started inferring was who was pregnant, who might become pregnant.”
Mark van Oppen 19:17
“somebody can just ask Ben what so-and-so salary is or somebody can just ask Ben why this person is on leave because he has access to my Slack.”
Anthony Enrico 21:45
“You know what helps a race car go very, very fast? Brakes.”
Anthony Enrico 23:47
“My favorite term to describe what is now in the hands of every employee is a spam cannon.”
Mark van Oppen 25:57
“There's this beautiful, very accurate, cleanly researched monologue that's very specific to me about why I should take a first meeting with them, but immediately I delete it because it smells like AI.”
Mark van Oppen 26:51
“You're asking them for an investment to read the next two lines”
Mark van Oppen 27:44
“What I don't want to do is truly cold approach them without some level of trust that can be corroborated from a perspective they care about.”
Mark van Oppen 29:38
“I really don't feel like there are too many hacks to growth anymore.”
Anthony Enrico 36:05
“my reaction was, oh man, that hides the lazy reps”
Mark van Oppen 38:43
“It took 30 minutes, cost $30, and he could have put that together over the course of two full days work maybe.”
Mark van Oppen 41:44
“you can't teach, give a shit. You can't teach somebody showing up with deep empathy for how this problem is affecting that customer.”
Mark van Oppen 43:12
“there's a huge difference between buying from a bot and buying from a human”
Anthony Enrico 48:55
Practical Advice

What should you actually do?

The playbook, split by the seat you sit in.

Revenue Executives

  • Assume AI is already in use across the company — desktop clients, uploaded files, MCP connections — and plan for visibility rather than prohibition.
  • Treat senior leaders who vibe-code as the highest-risk users: they carry the broadest access and tend to click through consent screens.
  • Give the CISO guardrails, audit logs, zero-trust controls and agent-level data redaction so the answer to AI adoption can be yes.
  • Ask whether every agent action can be traced to an owning person or team; if not, you cannot tell an agent's action from the employee's.

Founders

  • Run powerful personal agents on a dedicated or virtual machine, away from the laptop that holds founder-level credentials.
  • Before opening an agent to colleagues, check what its connections could reveal — salaries, reasons for leave, private Slack history.
  • Keep protected-category data, such as HR demographics, out of anything a planning or forecasting agent can reach.
  • Design how tools connect to each other, not just how they are monitored.

Sales Leaders

  • Use the over-personalised AI cold emails you receive as training material for what your team should not send.
  • Anchor first touches to corroborated trust: a customer anecdote, a common connection, a shared former employer or investor.
  • Make each message earn the next small investment — two more lines, a relevant white paper, then a meeting.
  • Use personal interests surfaced from public signals as a hook, without launching an essay that feels creepy.
  • Hire and coach for curiosity and empathy; reps who adapt how they work are pulling away, and empathy is what AI cannot supply.

RevOps Leaders

  • Issue agents downscoped, time-bound credentials instead of letting them inherit a user's access.
  • Report token consumption by agent, user, team and department so AI ROI policy is set on data.
  • Look for leftover services still consuming tokens with no active owner.
  • Put relationship-mapping tools in sellers' hands so leadership and investor networks are usable without asking executives to remember who they know.
  • Measure AI-assisted tasks by cost and time — Mark's rep reported $30 and 30 minutes for a list that would have taken about two days.

Marketing Leaders

  • Publish case studies, thought leadership and problem-specific documentation in a crawlable, indexable form so LLMs put you on the vendor list.
  • Pair indexed content with warm-intro programmes; the content lends credibility to the introduction.
  • Invest consistently and co-author with credible companies rather than AI-ghostwriting volume.
AI Takeaways

How AI actually changes GTM

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

The thesis

AI agents turned identity from a one-time check into a continuous question about every action, because each person can now delegate their access to a team of actors nobody vetted. The fix Mark describes is attribution — an identity per agent, downscoped and time-bound, owned by a human — which lets a CISO say yes instead of no. The same power tools have put a spam cannon on every desk, so the go-to-market advantage shifts back to trust, credibility and empathy.

Agent & automation ideas

  • An identity broker that issues each agent a downscoped, time-bound token for its objective, logs every action and alerts on unusual behaviour.
  • A token-spend view by agent, user, team and department that flags orphaned services and informs AI ROI policy.
  • Agent-level data redaction that strips protected categories such as HR demographics before a planning agent can use them.
  • A prospecting skill trained on won and lost deal dialogue that returns a territory target list with likely buying committees.
  • A relationship-mapping agent that surfaces warm-intro paths from leadership and investor networks, panels, filings and public commentary.
  • A shared team agent that answers each colleague only within that colleague's own permissions rather than its builder's.
Operations Takeaways

By function

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

Revenue Operations

  • .
  • .
  • .
  • .
  • .

Pipeline & Marketing Ops

  • .
  • .
  • .
  • .
  • .
  • .
Metrics Mentioned

The numbers, with context

$100M–$1B revenue; hundreds of employees
Where continuous authority surfaces first

Mark's description of the modern mid-size companies adopting AI tools without attribution, ahead of larger enterprises.

30 minutes, ~$30 in tokens
AI-built prospecting list

A rep's custom Claude skill produced a targeted territory list with likely buying committees; the token cost was reported using SecureAuth.

~2 full days of work
Manual equivalent

Mark's estimate of how long the rep would have taken to build the same list — around 50 accounts — and do the research by hand.

8,000 connections; ~100–200 introducible
LinkedIn connections versus real intros

Anthony's own network, illustrating why warm-intro discovery needs tooling rather than memory.

~2 years
Age of the agentic identity problem

Mark says the delegated-identity problem has really shown up in the last two years.

RevOps at 3 companies; LeanScale founded 2021
Anthony's background

Context Anthony gives for the Tom Miller metaphor and for how much RevOps work has changed since.

Entities

Companies, people & tools mentioned

Auto-extracted and linked into the knowledge graph.

Companies

SecureAuthCybersecurity

Mark's company, where he is CRO. It works on identity across workforce, customer and delegated non-human identities, and publishes an identity broker that gives each agent its own downscoped, time-bound, audited identity attributed to a human owner. Mark argues legacy identity vendors and non-human-identity point solutions each miss part of the problem.

00:12 · 10:56 · 11:43 · 41:33 · 46:19Company →
LeanScaleGTM Operations

Anthony's firm, started in 2021 doing mostly Salesforce and HubSpot RevOps work and now building agent fleets and agentic workflows. As founder and CEO he has God-mode access to its systems, which shaped his caution with his own agent, and he describes LeanScale's investment in podcasts, newsletters, white papers and case studies.

15:46 · 23:44 · 25:09 · 32:52Company →
Blue Box GroupCloud Infrastructure

Where Mark was primary seller before the company was acquired by IBM, per the episode show notes.

Company →
IBMEnterprise Technology

Acquirer of Blue Box Group, where Mark was primary seller, per the episode show notes.

Company →
HeptioCloud Native / Kubernetes

The startup founded by creators of Kubernetes, where Mark was the first go-to-market hire, per the episode show notes.

Company →
VMwareCloud Infrastructure / Virtualization

Acquired Heptio; Mark led a global sales team there after the acquisition, per the episode show notes.

Company →

People

Mark van Oppenguest

Chief Revenue OfficerProfile →
Anthony Enricohost

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

Co-FounderProfile →
Tom Millermentioned

Author of 'Call Your Shots' and a veteran go-to-market operator (multiple IPOs, buy- and sell-side M&A); Anthony Enrico's former CRO at Emailage.

Author of 'Call Your Shots'; former CRO, EmailageProfile →

Tools & software

ClaudeAI Assistant

Mark's example of an assistant creating agents behind the scenes while employees chat with it, and the basis of the custom Claude skill one of his reps built from won and lost deal dialogue to produce a targeted prospecting list. He also describes over-personalised cold email that reads like a Claude skill crawling public data.

Website ↗
Claude CodeAI Dev Tool

Named as one of the AI tools mid-size companies' software development departments are leaning into without yet having attribution for agent actions (transcribed as 'cloud code').

Website ↗
Model Context Protocol (MCP)AI Integration Protocol

Mark's example of how employees give AI context: a direct connection to a model context protocol server fronting a proprietary source of data.

Website ↗
OpenClawAI Agent

The agent Anthony experimented with and named Ben. Its power led him to run it on a dedicated or virtual machine, and its connections to finance, HR, project management, CRM and Slack showed how a shared agent could expose salaries or leave reasons.

Website ↗
SlackTeam Messaging

One of the systems Anthony connected to his agent, which meant colleagues could have asked it about information in his Slack.

Website ↗
VieuRelationship Intelligence

The prospecting tool SecureAuth uses to take leadership and investor networks out of executives' heads and give them to sellers: it maps LinkedIn connections, 10-Q filings, analyst and speaker panels, public commentary and interests into warm-intro paths to a target account.

Website ↗
LinkedInSocial Platform

The source crawled for the over-personalised AI cold emails Mark deletes, and one input to SecureAuth's relationship mapping. Anthony notes he has 8,000 connections but could make an intro to only one or two hundred.

Website ↗
GongRevenue Intelligence

The tool that turned Mark from a meeting-recording skeptic into a believer at a previous employer, once reps used it for customised follow-up and prospecting as well as summaries.

Website ↗
SalesforceCRM

With HubSpot, what most of Anthony's RevOps work meant when he started LeanScale in 2021.

Website ↗
HubSpotCRM

With Salesforce, what most of Anthony's RevOps work meant when he started LeanScale in 2021.

Website ↗
Methodologies referenced Token attribution by agent, user, team and departmentA prospecting skill trained on won and lost dealsRelationship-graph prospectingInterest-based hooks
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.

Agent access, attribution and guardrails are the operating questions behind putting AI to work across go-to-market, alongside the AI-built prospecting and relationship-mapping plays Mark describes.

Sales Enablement Build

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

Coaching reps away from spam-cannon outreach toward trust-anchored, incremental prospecting, and helping the curious sellers turn AI tools into a force multiplier.

Frequently Asked Questions

Straight answers

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

Why did AI agents break traditional identity security?

Traditional identity was a red-light, green-light decision: verify that a person is who they claim to be, then trust their judgment. Mark van Oppen, CRO of SecureAuth, argues that AI agents act with delegated free will, so every identity can now run a team of unvetted, unbounded, unsecured actors pursuing loosely defined objectives, often with shared credentials. The number of actors in an organisation rises sharply and the implicit human friction disappears, so security has to shift from a one-time check to assuring the right to act at every action layer on an ongoing basis.

What are the main ways AI agents cause security incidents?

Mark van Oppen names three vectors. Credential theft, where long-lived access tokens and API keys are stored inappropriately and passed around. Prompt injection, where an agent's instructions are taken over by a malicious or simply clueless actor and it goes off the rails. And rogue behaviour, where an agent pursuing an objective cuts corners, accesses information or takes actions nobody predicted. Together they move the goalposts from verifying identity once to verifying continuously that each action is permitted and scoped to its business purpose.

Can you tell whether an action was taken by an AI agent or by the person whose credentials it used?

Fundamentally, no. If an agent acts with a person's unique credentials, the action is attributed to that user identity. Mark van Oppen's answer is to stop agents inheriting credentials: SecureAuth publishes an identity broker that gives each agent its own unique identity with a downscoped, time-bound access token based on its objective, logs and alerts on every action, and ties the agent back to the human or team that owns it. That also makes token consumption visible by agent, user, team and department and exposes leftover services still consuming tokens.

What is continuous authority in identity security?

Continuous authority is the term SecureAuth uses for verifying, on every action, whether an identity — human or non-human — has permission to act and should have it, with access downscoped by default. Mark van Oppen sees it appearing first at modern mid-size companies with roughly $100 million to $1 billion in revenue and hundreds of employees, whose AI-assisted engineering teams do not yet have attribution for agent actions. He expects it to show up in ISO and SOC standards and in data-access and sovereignty controls.

How can a CISO enable AI adoption instead of blocking it?

Mark van Oppen says CISOs became the department of no because they understand the data-leakage and compliance risk while boards push for AI. Blocking does not work, since employees adopt tools whether or not they are sanctioned. The alternative is to green-light tools with guardrails: manage agent behaviour patterns, attribute every action to an audit log, keep a zero-trust posture across workforce, customer and non-human identities, stay compliant, and redact data at the agent level. That gives the CISO a visible, defensible map of what is happening and the ability to apply budget and data-security policy.

What can go wrong when senior executives vibe-code with AI agents?

An enterprise customer told Mark van Oppen its biggest security concern was vibe-coding C-suite members, who have far-reaching access and click through consent screens. In one case a senior executive at a fintech built a capacity model with an agent that could see HR demographics; forecasting workforce availability from dates, ages, gender, location and seasonality, it began inferring who was or might become pregnant, a protected category. The team removed the data and added controls. In another, an agent used by a read-only finance staffer correlated budget timing and line items and exposed an outgoing CEO's protected exit package.

Why do highly personalised AI cold emails fail?

Because the personalisation is inappropriate for the relationship. Mark van Oppen receives emails built from crawled LinkedIn history, even referencing where he went to college — accurate and well researched, but far too specific for a stranger — and deletes them because they smell like AI. He calls the tooling behind them a spam cannon. His guidance is to be specific but concise and to treat each message as asking for a small investment, such as reading the next two lines or opening a relevant white paper, before earning the next step.

How do you build B2B pipeline in 2026 if cold outreach is not working?

SecureAuth anchors pipeline to trust. Mark van Oppen's team starts from customer anecdotes, warm introductions and credible voices, and uses a relationship-mapping tool called Vieu to find paths into target accounts through leadership and investor networks, panels, filings, public commentary and shared interests. To scale it, they publish case studies, thought leadership and problem-specific documentation that LLMs can crawl and index, so the company appears when buyers ask for vendors and warm intros carry more credibility. Host Anthony Enrico's summary is that there are no hacks to growth left.

Do AI sales tools help weak reps or strong reps more?

Both, according to Mark van Oppen. He was a skeptic who worried meeting recorders would hide lazy reps, but after Gong was rolled out at a previous employer, reps used it for customised follow-up and prospecting and could still defend their deals, which raised the floor. The best reps go further by automating manual work: one built a custom Claude skill from won and lost deal dialogue and produced a territory target list with likely buying committees in 30 minutes for about $30 of tokens, instead of roughly two days by hand. He believes curious sellers who keep changing how they work will be the most successful, and that empathy remains the part AI cannot replace.

Timeline

Jump through the conversation

Click any chapter to open it in the transcript.

00:00Cold open + intro 02:03Identity used to be red light, green light 03:44What your employees are already doing without telling you 07:05Three vectors: credential theft, prompt injection, rogue behavior 08:36The CISO squeeze: becoming the department of no 13:28Continuous authority — and who feels it first 17:03The vibe-coding C-suite and the pregnancy forecast 20:23The agent that exposed a CEO's exit package 25:00Every employee now has a spam cannon 28:26Is cold email dead? Building pipeline on trust 31:10AEO, indexed content, and scaling the warm intro 38:21Does AI raise the floor or 10x your best reps? 40:59The $30 prospecting list 43:13You can't teach give a shit
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 If you can focus and deeply root your actions in solving that customer pain, then you're

0:08 going to do great.

0:09 And that's what the best sellers do.

0:12 My guest today is Mark Van Oppen, Chief Revenue Officer at SecureOff, a go-to-market leader

0:18 who has spent his career selling the least optional technology on earth.

0:23 It's really fun to speak to somebody who's like-minded around the way we approach the

0:27 industry.

0:29 If an agent is using my credentials to execute a task, how is it even technically possible

0:34 to understand if that action is being taken place by an agent versus Anthony?

0:40 Fundamentally, it's not, right?

0:42 If an agent is acting with your unique credentials, I would attribute that action to the credentials

0:47 that are being used, right, to the user identity that's taking that action.

0:53 What are the best ways to build pipeline in 2026 and, you know, is the email dead?

1:00 Do we have to do other things?

1:03 What is filling up your pipeline or the pipeline of your peers in similar roles?

1:08 Yeah, we're taking a strategy of trust.

1:11 So how do you get trust?

1:14 You have to have a customer anecdote.

1:17 You have to have a warm intro.

1:19 You have to have somebody to say, "I think highly of this technology, and I'm a trusted

1:26 voice that will help you earn a first meeting with a prospect."

1:32 I mean, you really, really have to do that, and that's the hard way to do it.

1:38 I really don't feel like there are too many hacks to growth anymore.

1:50 Mark, you've said identity used to be a red light, green light decision, Anthony is Anthony,

1:56 all done.

1:57 What exactly broke that model the day we started handing real tasks over to AI agents?

02:03Identity used to be red light, green light

2:04 Well, it's a great framing to start this conversation.

2:08 So the first thing that changed was the implicit trust and human-level friction that existed

2:15 naturally.

2:16 I could verify Anthony is Anthony, and I can trust Anthony's judgment as the employee that

2:22 I hired or the customer that is a paying trusted identity that is getting access to my system.

2:30 There was a lot more sort of understanding of who's acting and when and why.

2:35 Now, the emergence of AI as something with a delegated free will, for lack of a better

2:41 term, has created every identity, the potential to run a team of unvetted, unbounded, unsecured

2:50 actors that make decisions based on a loosely defined objective, and that creates a level

2:57 of risk that is exponential in your organizations.

3:01 So not only is an exponentially higher number of unique actors in all of these non-human

3:06 identities that are running around in your organization, but there's credential sharing,

3:10 there is an intrinsic change in what you can trust and who you are giving access to.

3:17 So instead of just giving top-level access to Anthony, I have to pivot and look at assuring

3:25 the right to act at every action layer in an ongoing way.

3:28 Does that make sense?

3:30 It does make sense.

3:31 I'm wondering if there are certain things that people are doing that they don't even

3:36 realize is opening up the level of risk.

3:39 Are there some examples that you're seeing people do right now that's just totally leaving

3:43 them open?

03:44What your employees are already doing without telling you

3:44 Yep.

3:45 So I think about it in a work context, where whether you are selling genes on the internet

3:53 or you are managing a software as a service, your employees are starting to tinker with

4:01 AI, meaning they've downloaded a desktop client.

4:06 They've started to use AI to formulate emails or to write a little more eloquently or on

4:13 target.

4:14 And no matter how they've set up and started interacting with this new tool that's available

4:20 to them, they're giving it context to get better output.

4:24 And that context could be uploaded files.

4:26 It could be a direct connection to an MCP server or model context protocol server that

4:32 is fronting some proprietary source of data and people don't realize when they're just

4:37 bantering back and forth with Claude, for example, Claude is creating agents to pursue

4:43 and get answers on your behalf or to create the document that you've asked for.

4:49 It is creating and taking the access that it knows about you to go execute some action.

4:56 And the more things you connect it to, sanctioned or unsanctioned, the wider and the more ungoverned

5:01 and unaware the enterprise security controls are to that behavior pattern.

5:07 So if you think about it like the concentric circles of visibility, the most center trusted

5:14 actor is a direct action that is Mark or Anthony and the keys I've given Mark or Anthony to

5:19 take actions that he or she is doing every day.

5:24 But then you start to watch an agent get delegated access pursuing some objective and that agent

5:29 might spawn subagents.

5:31 You might have a recurring or scheduled actor starting to happen, but it's happening whether

5:36 you've sanctioned it or not.

5:38 And I don't know of a business that can completely say there is no AI in our organization because

5:46 people are people and they're going to start using the shiny new toys that are available

5:49 to them.

5:50 And whether it's sanctioned or not, they think they're doing something harmless.

5:53 But all of a sudden it is opening up a vector of risk that the vast majority of businesses

5:59 just aren't aware of yet.

6:01 And I think we're going to see a wave of alerts and news articles around how this is going

6:08 to bite people over time.

6:10 Yeah.

6:11 And I think we have started to see some of this already.

6:15 We saw the open AI agents that were used to hack into the hugging phase and we're seeing

6:21 a couple other kind of large scale.

6:24 I feel like we've needed those big events to start ringing some of the alarm bells because

6:30 a lot of people are just using this as if there's no risk at all and it's like this

6:34 magic wand that you can use with no consequence.

6:37 But now that some of those have started to happen, I anticipate there's going to be way

6:40 more, especially as the models have gotten more powerful.

6:43 I think securing this stuff down is going to become the number one problem for a large

6:49 enterprise.

6:50 Yeah, I firmly agree.

6:53 And there's really three inherent vectors that vulnerability can surface.

7:01 It's credential theft, so a lack of awareness of who aware long lived access tokens are

07:05Three vectors: credential theft, prompt injection, rogue behavior

7:08 being stored or API keys or whatever the case may be.

7:12 Those are getting handed off or stored inappropriately and they're just being passed around.

7:17 So credential theft is the first one.

7:20 Then there's prompt injection, right, you're seeing the instructions to this agent getting

7:26 taken over by some malicious actor or even just clueless actor where the intended purpose

7:34 goes rogue in some way or goes off the rails in an unintended pattern.

7:38 And then just straight up rogue behavior where an agent is pursuing an objective and it cuts

7:45 corners or accesses information or starts querying or taking action in a way that you

7:50 can't predict.

7:52 So you kind of have those three ways that this will manifest and it's changed the goalposts

7:58 that we're all aiming for, that it's not verifying you are who you say you are, it's verifying

8:04 in an ongoing and continuous way.

8:06 Do you have the right, do you have the permission and is that action appropriately scoped for

8:12 the business purpose behind it?

8:14 And that's an ongoing sort of omnipresent posture that you're going to have to maintain

8:18 in perpetuity with this new world of power tools.

8:22 Yeah, and obviously there's so much pressure to leverage AI to increase productivity, effectiveness,

8:30 efficiency.

8:31 There's board level mandates to have these AI initiatives really kicked off and invested

08:36The CISO squeeze: becoming the department of no

8:39 heavily into.

8:40 So I think you have those two opposing forces and one role this tends to put in a really

8:46 tough spot is the CISO in an organization.

8:49 And in a world where there's a heightened level of anxiety to get productivity out of

8:54 these super tools, you don't want to be the person that has to say no to everything.

9:01 How should that person be thinking about this and how do you get them in a position where

9:07 you can give people the tools they need, but also have the security that protects the company

9:13 at all, at a whole?

9:15 Yeah, so I think you're touching on a broader theme that we've seen in many, many of our

9:21 customers already.

9:22 The reason they're not all systems go adopting AI without any concerns is that this risk

9:30 is understood by a CISO or by somebody who's thinking about compliance and data leakage

9:37 and they're worried about the idea of what happens if we just say green light across

9:43 all of AI adoption, what happens to our proprietary information, the information that you're now

9:51 making accessible to these different models.

9:54 And that's put the CISO in a position of being the department of no, just a big old stop

10:01 sign around driving AI adoption.

10:04 And what we endeavor to do is allow that CISO to be the department of yes.

10:10 We want to help them green light the adoption of a variety of tools.

10:15 And like I said before, it's happening whether it's sanctioned or not.

10:19 So we want to help you get guardrails, be able to deal with the workforce identity,

10:24 be able to deal with the customer identity and be able to deal with all these non-human

10:29 delegated identities from the different humans out there, right?

10:32 The humans aren't going anywhere, but the agents are also there.

10:36 And so can you manage a behavior pattern?

10:39 Can you attribute every action to an audit log?

10:42 Can you maintain a zero trust posture throughout all the different identity types?

10:48 And can you do so in a way that keeps you compliant?

10:51 Can you do data redaction at the agent level?

10:56 And the answer is yes with Securoth, but it's not a one-trick pony.

11:01 It's not one product to solve this problem.

11:04 It's making sure that you have a holistic approach across human and non-human.

11:09 And you have the credibility and proven scale to do this for business critical customers

11:15 at the largest scale.

11:18 If an agent is using my credentials to execute a task, how is it even technically possible

11:24 to understand if that action is being taken place by an agent versus Anthony?

11:30 Fundamentally it's not, right?

11:32 If an agent is acting with your unique credentials, I would attribute that action to the credentials

11:37 that are being used, right?

11:39 To the user identity that's taking that action.

11:43 So what Securoth does with these non-human identities, right, these agentic workloads

11:49 is actually publishes an identity broker.

11:53 So every agent that is acting on Anthony's behalf doesn't just inherit Anthony's credentials.

11:59 It gets its own unique identity, a downscoped access token based on the objective it's trying

12:03 to solve, much more time bound access, and every action is attributed to an audit log.

12:11 And then there's alerting around what is being done.

12:15 So you have an attributable history across every action being taken.

12:20 You can look at like how many tokens were consumed by agent, by user, by team, by department,

12:26 and you can look at and set policy around the ROI of certain adoption of these tools.

12:33 But you have a level of visibility and control that's rooted in assigning a unique identity

12:38 to the agent itself, and that unique identity, that agent, is attributed to Anthony.

12:45 So Anthony's dozens of agents that are pursuing lovely tasks that are all well-intended are

12:51 owned by a human or by a team, right, and you have an ability to tie that logic together

12:57 so that you can attribute maybe a leftover service that's just sitting there and churning

13:02 through tokens.

13:04 Are there any industries you think this is going to be particularly important for, or

13:09 do you think this should be a ubiquitous standard that any enterprise takes on?

13:14 Yeah, I think the term we've started using is continuous authority, meaning you have

13:19 a continuous authority verification attributed to every action, right?

13:25 Does this person have the rights to do it?

13:27 Does this identity, whether it's human or non-human, have permission?

13:28Continuous authority — and who feels it first

13:32 And should they have permission to do it?

13:33 So everything's downscoped, and I think that's the posture that everybody will need to maintain.

13:38 I think that's just the evolution of where identity security is going.

13:44 But it's starting to happen, and it's starting to come to the fore in really sort of midsize

13:53 companies first, right?

13:54 Think of it, maybe $100 million to $1 billion in revenue.

13:59 They have hundreds of employees, but maybe not thousands.

14:03 And they are modern enough that they're starting to really adopt these tools.

14:10 Lots of the software development departments are AI-assisted, and they're leaning into

14:17 cloud code.

14:18 They're leaning into these tools, and they don't yet have attribution.

14:22 That's a pretty sophisticated audience, so they haven't yet started running into sharp

14:26 corners in the dark, getting bit by these problems yet, but they will.

14:31 And right now, there's a really big miss in that audit log, and being able to verify,

14:38 is it Anthony taking action, or is it some delegated actor on Anthony's behalf that is

14:43 taking that action?

14:45 And eventually, I think that's going to show up in compliance standards, in ISO standards,

14:50 in SOC standards, in all kinds of sort of data access and sovereignty controls.

14:57 People are going to worry about the delegated permissions that are existing in this sort

15:03 of ungoverned pool of free will and judgment that didn't used to exist.

15:07 It's effectively a sub-employee that you have a background check to that is starting to

15:15 take action with your data, so it'll have to be treated in a really first-class security

15:20 way.

15:21 Yeah, I think the moment I started to get a little nervous was when I was experimenting

15:27 with Open Gla, and making sure, hey, oh wow, this is so powerful, I need to have a dedicated

15:33 machine, or at least a virtual machine that needs to be off my computer, because it can

15:36 go into anything.

15:39 And I named him Ben, and Ben did a lot of fantastic work for me.

15:44 But I think that was the first time.

15:46 And sitting in a founder's CEO seat, I do have God mode access to everything at Lean

15:51 Scale.

15:52 I can see the credentials, I can see the HR system, I can see all of our customers' data,

15:58 I can see anything.

16:00 So my credentials passed down to AI agents.

16:07 Pretty alarming.

16:08 And could potentially create artifacts that end up being public, or it's connected to

16:15 my email, it's connected to my Slack, it could accidentally pull some data over here and

16:19 Slack it to someone over there, so we have had to be very intentional about our security

16:23 on AI.

16:26 Not only having the right tools to monitor things, but also just structurally how we

16:31 build and connect the different tools that we're using.

16:34 So I think it's a massive, massive problem.

16:38 And also, if you do it right, can unlock the productivity in a real way.

16:45 But if done wrong, the risks are pretty severe.

16:48 Yeah, I totally agree, and I think it's a good example to talk about OpenClaw and talk

16:54 about the varying sophistication of a user.

16:57 Because you are aware enough to be worried that you're delegating this kind of access.

17:03The vibe-coding C-suite and the pregnancy forecast

17:05 But we had an enterprise customer at a significant scale joke that their biggest security concern

17:14 is vibe-coding C-suite members.

17:17 People that are very senior level that have really significant asks who are starting to

17:22 vibe-code and create forecast models, but they're just clicking through consensus and

17:26 giving access to really significant reaching identities.

17:33 And we had one start building a forecast model that started predicting whether or not the

17:39 women in the company were pregnant based on capacity, and it's like, "Okay, you can't

17:42 do that."

17:44 There's limitations around the data you can and can't use.

17:48 And of course, an agent doesn't have that judgment, doesn't have that human context,

17:53 doesn't have the sense of PII or not, unless it's been trained in a very anomalistic way.

18:00 And so you start to think about the skepticism and the sort of critical eye of the operator,

18:09 and they have to respect the fact.

18:10 They have this massively powerful risk generator that could also really make them this force-multiplying

18:17 worker on your team.

18:19 But there are some foot guns to be aware of.

18:23 Hang on.

18:24 I have to double click on that story.

18:25 First of all, is that real?

18:27 They built a forecasting model that was predicting what women in the company would get pregnant,

18:33 and then two, what executive was that?

18:36 Were they in sales?

18:37 Were they in...

18:38 I hope they weren't in HR.

18:39 No.

18:40 There was a senior executive in a fintech, and they were thinking about capacity models

18:46 of their team.

18:47 They were talking about productivity across departments, and it was a complicated forecasting

18:50 model.

18:51 But one of the things they had access to was an HR system, a couple of the demographics

18:58 of their workforce was one of the pieces they had information about.

19:02 And part of the reasonable actions that the agent was taking was forecasting availability

19:09 of the workforce.

19:10 And okay, they took date ranges, they took age ranges, they took gender, they took location,

19:15 they took seasonality.

19:17 And one of the things it started inferring was who was pregnant, who might become pregnant.

19:23 And obviously, that's not information you can make a capacity planning decision around,

19:30 and because it's a protected category, reasonably so.

19:34 And they did the right thing.

19:35 As soon as they realized that was happening, they omitted the data, they added a few more

19:39 controls, but it's an example of giving an agent or giving a model, an objective, reasonable

19:46 forecast planning, but the absence of human judgment around what can and can't be used

19:52 in that discussion or in that exercise.

19:56 Right.

19:58 That's absolutely insane.

19:59 And I think a really good example, especially for senior executives that do have very, very

20:05 high access, and they may not be realizing what they're delegating down, I think.

20:10 Yeah.

20:11 Another one that was an interesting trait that popped up that was not quite the C-level

20:16 problem.

20:17 It was a read-only staffer in the finance department inadvertently correlated and exposed

20:22 the exit package of an outgoing CEO, which of course is legally protected and all kinds

20:23The agent that exposed a CEO's exit package

20:29 of non-disclosure protection there, but the timing, the line item in a budget, and the

20:38 sort of association of the information that was accessed by this agent that, again, was

20:44 looking at budgets and forecasts and capacity kind of stuff, it put two and two pieces together

20:49 that a human wouldn't necessarily have been able to do or wouldn't have done it because

20:53 it's that would have been reaching pretty far beyond, but wide-reaching information sets

20:59 that they can then be sort of correlated, exposed something that was supposed to be

21:04 private.

21:05 Yeah.

21:06 And especially in a lot of these cases, people are trying to get to the point where you can

21:10 get to multiplayer mode with an agent.

21:12 So the agent-based employee that many people can interact with, where it has a set of skills,

21:21 it has its own connection to certain systems, and I was kind of using Ben that way.

21:26 So Ben, I had him connected to the finance systems, I had him connected to our HR system,

21:32 I had him connected to our PM system and our CRM, so he had access to everything.

21:38 And people started using him for certain things like, "Oh, we want to do a forecast report

21:42 over here.

21:43 We can use XYZ over there."

21:45 And then I quickly realized, somebody can just ask Ben what so-and-so salary is or

21:51 somebody can just ask Ben why this person is on leave because he has access to my Slack.

21:57 And as soon as we kind of opened it up in that way, then it was like, "Okay, we got

22:01 to shut this down and reel it back."

22:04 Exactly.

22:05 And that's the sort of well-intentioned effort that starts to expose things that you didn't

22:11 initially consider.

22:14 And that skeptical sort of zero trust posture is the responsibility of a CSO to maintain.

22:22 And so they're put in this uncomfortable position to be the Department of No and saying, "I

22:26 need to limit the use of this because, oh my goodness, there are so many downstream

22:31 effects of the widespread adoption of AI."

22:34 But at the same time, the board, the investors, the employees are all clamoring to use these

22:39 tools and lean into them and they're moving so fast that the power of these tools is so

22:45 alluring.

22:46 And so they're getting adopted and they're getting used and people don't quite realize

22:52 it's a harmless use of this.

22:54 Even though it's not sanctioned, I'm going to start using this account and I'm going

22:57 to start using this tool.

22:58 And I'm not really connecting it to systems, so it's not that egregious.

23:02 But you start to see this waterfall effect of it's happening whether you like it or not.

23:08 And I'd much rather arm a CSO with a really strong, auditable, defendable and visible

23:15 map of what's happening in their organization and give them the tools to apply policy and

23:23 apply not only budget policy, but data reduction and data security policy to the adoption of

23:29 these tools.

23:30 And it's been a lot of fun to start offering that to CSOs in these enterprises that are

23:37 feeling it for the first time.

23:39 Yeah, I think those guardrails are really important.

23:44 Before starting the Lean scale, I ran RevOps for three companies and one of the CROs I

23:47 worked with, he would always say, "You know what helps a race car go very, very fast?

23:55 Breaks."

23:57 Because if you don't have breaks, then you can't just let the engine run and open it

24:02 up as much as you want.

24:03 You have to be able to stop yourself from getting in a crash.

24:07 So I think in a lot of ways, it's this too, it's like, "Hey, yes, we can be hyper productive,

24:11 but we got to put those guardrails in place first."

24:13 I'm curious.

24:14 Yeah, I really like that phrase.

24:15 I'm totally going to steal that because that's a great example of something that's nonintuitive

24:22 and a counter control that allows you to move fast with confidence.

24:26 Well, his name is Tom Miller.

24:29 He wrote the book, Call Your Shots.

24:31 That's an excellent go-to-market book and he has plenty of colorful metaphors and analogies

24:38 that I steal all the time.

24:40 So I think you have to pay him a royalty, but it should be fine.

24:43 All right, I'll be sure to look it up.

24:46 So I know you're sitting in a CRO role today.

24:51 AI has impacted go-to-market quite a bit in pretty much every aspect.

24:57 It's completely changed the dynamic of how you get in front of customers, how you show

25:00Every employee now has a spam cannon

25:04 up in the market, and a lot of things are changing.

25:08 It's changing so quickly.

25:09 I started LeanScale in 2021.

25:11 We did RevOps.

25:12 Mostly it meant some Salesforce and HubSpot work.

25:16 Now we're building full agent fleets for companies.

25:20 We're building agentic workflows for companies.

25:24 How are you leveraging AI in a go-to-market context?

25:30 We're leveraging AI in a lot of ways, but I think one of the things that's worth just

25:39 noting is that power tools have emerged for BDRs, for sales reps, for RevOps teams, for

25:49 people like me that want to do pipeline inspection, want to do deep analysis of the health of

25:54 deals.

25:57 My favorite term to describe what is now in the hands of every employee is a spam cannon.

26:04 That spam cannon phenomenon is super dangerous unless you have a sense of skepticism around

26:12 it.

26:13 I get cold emails from people soliciting me all the time to sell some various product

26:21 that are way too customized for the level of stranger that I am to them.

26:27 This is not a warm intro.

26:29 This is somebody that's crawled social platforms like LinkedIn.

26:34 They've looked at my history.

26:36 They'll maybe make a reference to where I went to college or something, but it's a level

26:41 of personalization that would be inappropriate for a BDR to do about me, and it sticks out

26:50 like a sore thumb.

26:51 There's this beautiful, very accurate, cleanly researched monologue that's very specific

26:57 to me about why I should take a first meeting with them, but immediately I delete it because

27:02 it smells like AI.

27:03 It smells like something that is a clod skill to crawl all the public data points they can

27:10 find about me.

27:12 I use that as an example to warn my team about what not to do when prospecting.

27:19 We're trying to earn the next engagement, earn the next trust, and that behavior has

27:26 shown up across all kinds of BDRs where you're sort of too specific when you're reaching

27:33 out to somebody and it's clearly inappropriate for the level of relationship you have.

27:39 Maybe try to add value and be specific but concise.

27:44 You're asking them for an investment to read the next two lines, and then you're asking

27:48 for an investment to maybe open this white paper that's specific to them and matters

27:53 to their persona, but you're trying to earn just little by little each further investment

28:01 in your communication, especially when it's cold.

28:05 Does that answer the question or should I talk about some of the tooling or how we're

28:09 using AI among our staff itself?

28:11 No, I think we can get into that too, but one question I have, what are the best ways

28:18 to build pipeline in 2026?

28:24 Is email dead?

28:25 Do we have to do other things?

28:26Is cold email dead? Building pipeline on trust

28:27 What is filling up your pipeline or the pipeline of your peers in similar roles?

28:32 Yeah, we're taking a strategy of trust, so how do you get trust?

28:38 You have to have a customer anecdote.

28:41 You have to have a warm intro.

28:43 You have to have somebody to say, "I think highly of this technology and I'm a trusted

28:50 voice that will help you earn a first meeting with a prospect."

28:56 So we're using things like a prospecting tool that looks at all of LinkedIn connections

29:03 and 10Q posts from public companies and seeing what they're saying, maybe analyst panels

29:13 and speaker panels.

29:14 We're looking at all this public information and trying to correlate who we might know

29:19 so we can get a warm door opening and handshake there.

29:24 Maybe there's common history in their employment, maybe there's a warm intro among our investors

29:31 or their investors or some mix there, but it's establishing trust early.

29:38 What I don't want to do is truly cold approach them without some level of trust that can

29:45 be corroborated from a perspective they care about.

29:48 Whether it's me, whether it's a common connection point, whether it's a common former employer

29:54 or something like that, there's got to be some trust earned and established.

29:59 So that's really the nucleus of how we're trying to approach new logos.

30:04 Anchor to trust somehow.

30:06 Yeah, I think that's always the number one way to get in front of a prospect to get referred

30:12 in by a peer or have them see a company leveraging your product as well.

30:26 The only thing I think about in that context is can it go fast enough?

30:30 These companies that are, you just raise your Series A, you raise your Series B, there's

30:34 quite a bit of pressure to grow quickly and keep up with momentum and you likely made

30:39 a lot of promises or in the fundraising process too.

30:42 So can you, although it's effective, can you scale it and can you accelerate it and can

30:50 you inorganically move it forward fast enough to keep up with the growth that you have to

30:54 achieve?

30:55 Yes, and I think that comes from credibility.

31:01 So you're familiar with SEO and have you heard and started talking about AEO?

31:05 Oh yes.

31:06 Yeah.

31:07 We've had a couple of platforms that...

31:10AEO, indexed content, and scaling the warm intro

31:10 We've had a couple of platforms that help companies optimize for AEO and yeah, a couple

31:16 of AEO engineers, so absolutely.

31:20 Yeah.

31:21 So I think the way to scale that is thought leadership pieces, the case studies, robust

31:30 documentation around specific problem statements that are written in a pattern that's crawlable

31:36 and searchable and indexable and that will help you show up.

31:40 When somebody starts asking an LLM how to solve this problem, they're using phrases,

31:47 they're using terms, they're framing this and you have to have written content that

31:52 can be picked up and that can be indexed by those.

31:55 Otherwise, you're not on the list of potential vendors for this problem and then you magnify

32:01 that with the best sort of warm intro indexing possible.

32:07 And that takes a lot of effort, but there are some really cool tools out there that

32:10 help you magnify and take advantage of your executive team's extended network, your investor's

32:16 extended network and everybody else in the company's extended network where you may be

32:20 able to earn a warm intro that then gets bolstered by the credibility of the sort of indexed thought

32:27 leadership pieces that are out there.

32:29 But it makes it just a little less cold when you're reaching out to a target account.

32:35 Absolutely.

32:36 No, and I think you have to have some proprietary content, you have to have some opinionated

32:44 takes on something and then you have to show up on other people's sites as well and be

32:48 mentioned in other areas and it takes a lot of investment.

32:52 So I know we talk a lot about, you know, LeanScale invests quite a bit in content.

32:57 We do podcasts, we do educational content, we do newsletters, we do white papers, case

33:01 studies, the whole thing.

33:03 And it takes a long time, but once you have been doing it consistently, it really starts

33:08 to compound and it's a tough thing to replicate and keep up and you can't do it through just

33:16 AI ghost writing a million articles about something like you actually have to get credible

33:22 companies and sources to also co-author content with you.

33:28 So it's a huge moat if you invest in it.

33:31 Are there any tools that you're using that are helping with this, especially you mentioned

33:38 LinkedIn mining connections, because I think that's an interesting one where, yes, I have

33:43 8,000 connections on LinkedIn, but I probably really only know and could make an intro to

33:50 about 1 to 200 of them.

33:54 How are you navigating that?

33:55 Yeah, we use a tool called Vue.

33:58 So V-I-E-U and it's been helpful for us to just take the monetization of our extended

34:11 leadership team and investor team's network out of their hands and put it in the hands

34:16 of a seller.

34:17 So a seller can say, "Hey, I want to go after Acme Company.

34:20 Who do I know but Acme Company?"

34:22 He or she can put this information into Vue and Vue can say, "Okay, well, you have several

34:29 paths to a secondary connection and there was a common thread or somebody who knows

34:35 this board member spoke on a panel together and had this thread or this connection point,"

34:42 or "They had a breach last year.

34:44 This was the attributed fault.

34:46 This is somebody who gave commentary on the article that wrote about it and they're connected

34:53 to your company in this way."

34:55 So it'll really map all the publicly available sources and allows you to seek that warmer

35:02 intro where it's not definitively saying, "I, Anthony, have a first party connection

35:07 that I would remember."

35:08 It's not really for those 200 connections that you feel good about introducing.

35:12 It's to actually do something useful with a whole bunch of data points that are more

35:18 complex and nuanced and complicated than you would know first person about yourself and

35:23 putting the action to mine that context and that data in the hands of my sellers instead

35:28 of counting on you to just remember the correlation or a connection across everybody you may have

35:35 connected to over the last 15 or 20 years.

35:38 Yeah, and I think there's just so much data to be going through, and I think if you're

35:43 manually trying to go through your LinkedIn or manually trying to go through your connections,

35:46 it's really, really difficult.

35:48 But knowing that that's the key, like any edge you can get into the warm intro space,

35:55 of course also backed by content and searchability, I mean, you really, really have to do that.

36:02 And that's the hard way to do it.

36:05 I really don't feel like there are too many hacks to growth anymore.

36:11 It's like you have to do this the good old fashioned hard way and compete really, really

36:17 heavily because everybody is weaponized with all of these AI capabilities too.

36:21 Yeah, very, very much so.

36:23 And one of the things that's important is that it's not just looking at LinkedIn data.

36:29 It's looking at interests, right?

36:31 Maybe there was a comment made publicly on Rossignol, a brand that makes skis.

36:38 And so you know this person that happens to be a CTO at Acme company or even a CISO at

36:44 Acme company, it likes or follows a social account of a ski manufacturer and has interacted

36:52 with their content about excited to use this new product that launched last fall.

36:57 You start to associate that person has an interest in skiing.

37:02 I can use that as a way to set a first meeting because you know that that's a kind of an

37:09 uncommon level of depth and correlation.

37:13 And so what Vue does and what we're trying to teach our team to do is use all available

37:19 context you can to put together a really hard set of data points about your target customer

37:27 so that you can be super specific.

37:30 But you do it in a way where you're not going to immediately launch an essay at this target

37:36 that feels creepy and too deep.

37:41 You want to do it where you're offering them something of value.

37:45 You're hooking them with a direct reference to something you know they're interested in

37:49 based on the context you've put together and then earning a little further investment,

37:53 every one of those interactions, so that you can get to a meeting, ultimately a meeting

37:58 where then you earn a demo and you earn a pilot or an investment in your desired outcome.

38:07 Are you seeing all these tools kind of raise the floor of your current reps or is it your

38:15 really good reps that are taking this and just 10X'ing their productivity?

38:21Does AI raise the floor or 10x your best reps?

38:21 It's a little bit of both.

38:22 And I have to share that I was an AI and a meeting recording skeptic, right?

38:31 The first sort of tools that showed up were things like meeting recorders and automated

38:37 notes and summarized action items that took the transcript of these meetings.

38:43 And my reaction was, oh man, that hides the lazy reps because all of a sudden they're

38:48 not actually taking notes in real time and capturing the follow-on items.

38:53 It's getting automated for them and I actually don't know.

38:56 Are they good at doing this themselves?

39:00 Do they have command of this or is it all sort of hidden by automation?

39:04 Now is that a bad thing?

39:06 Maybe not.

39:07 It's in raising the floor of your more mediocre reps.

39:12 But what I started to see was that it also turbocharged your most capable reps.

39:18 And I became a believer after we actually installed and started using Gong at my previous

39:26 employer and a lot of the reps started leaning in to not only the initial summarization and

39:33 action items that it was surfacing in that tool, but they started using it to build customized

39:40 material.

39:41 They were starting to use it for custom follow-up, for prospecting, for all kinds of more detail.

39:49 And if they brought a level of skepticism, it didn't come across as creepy.

39:54 So my initial concern was that it would hide my low performers and make it harder to tell

40:00 who really had command of their deals.

40:04 But how it's borne out is people leaned in to the tools and started using it and all

40:10 of a sudden they're able to accomplish their business and they know their business well

40:17 enough to defend it in conversation, which gives me a lot of comfort that they are deeply

40:22 involved.

40:23 They're just using power tools that didn't exist when I was in that role.

40:28 What did it do to your good reps?

40:29 So it sounds like, okay, help some of the, we'll call them poor performers, catch up

40:34 to maybe what the good ones were already doing.

40:37 Are they even or are your good ones taking it to another level?

40:42 The good ones are leaning into ways that they can automate manual tasks.

40:49 So I have one guy who started building a prospecting list and he could have done this manually.

40:59The $30 prospecting list

41:00 He could have built this target list of 50 accounts or whatever it was over the course

41:05 of a day and done all this research himself, but he took the initiative to build a custom

41:10 Claude skill, gave it, started training it and giving it context of here's an example

41:16 of all the dialogue I had in a deal that we lost.

41:18 Here's one that we won.

41:19 Here are the traits that mattered.

41:20 Can we find a list that were in, you know, these parameters that have might have similar

41:25 pain points.

41:26 And all of a sudden he had a targeted list in his territory with a likely buying committee

41:33 identified and he was able using secure off to send me the cost of tokens that it burned

41:40 to do this task.

41:42 And it was like $30, right?

41:44 Like it was, it took 30 minutes cost $30 and he could have put that together over the course

41:51 of two full days work maybe.

41:54 But I was thinking, oh my goodness, if you're willing to lean into these tools available

41:59 to you, you can make a decision and be a force multiplier for your very best reps when it

42:04 comes to sort of being a targeting system for your prospecting efforts.

42:09 And he started doing that.

42:10 It was really fun.

42:11 So I think these AI tools were just scratching the surface of how they become a force multiplier

42:21 for all kinds of roles and the sellers that are curious, that are willing to learn that

42:26 are not stagnant in the way they work are going to be the most successful and we're

42:34 seeing it in real time in our organization.

42:39 Is there anything in your opinion that cannot be facilitated or done by AI that, hey, this

42:48 is where the sales team really holds their ground and adds their value where AI is not

42:55 replacing that?

42:57 I think it comes down to empathy, right?

43:00 Empathy is something that can't be replaced.

43:03 I was speaking to a previous colleague recently and we used the sort of the casual term of

43:12 you can't teach, give a shit.

43:13You can't teach give a shit

43:14 You can't teach somebody showing up with deep empathy for how this problem is affecting

43:21 that customer.

43:22 And AI is clinical in how it diagnoses information and spits out a result, right?

43:31 A really good seller can look at how a problem is manifesting in a specific customer and

43:38 say, okay, I'm going to tailor my response to be an extension of their team and focus

43:44 on the business outcome that matters to them.

43:47 And if that works out, the byproduct of that is using our product, excellent, then they

43:52 pull you across the line.

43:54 But that sort of human judgment and empathy is not something that AI will be able to replace.

44:01 I think it takes a level of earnest care from a seller to anchor to a business problem and

44:09 how it's actually hurting a specific customer because they all are a little different, right?

44:14 All these businesses are a little different.

44:16 All these people's jobs are a little different.

44:18 And if you can focus and deeply root your actions in solving that customer pain, then

44:26 you're going to do great.

44:28 And that's what the best sellers do.

44:30 Yeah, and I think definitionally, if you have a friend who's listening to something tough

44:35 you're going through, having a human hear it is the part that's adding value and empathize

44:44 with you, as you mentioned, versus I could get some text back, but it doesn't count because

44:49 it's not a human that is really taking some of that burden on.

44:54 Because I think when you're still selling to people, people are people, and you have

45:01 to build that empathy and they have to feel heard.

45:04 They have to feel like, I'm stealing that one by the way, you can't teach give a shit.

45:08 So that's my new, I'll trade you Tom Miller's quotes.

45:14 But that in essence is what makes it so important and what builds that trust.

45:20 And then also you're on the hook for it, you're accountable.

45:25 Somebody on the other side, if something goes wrong there's potential justice to be served.

45:30 So I think all those components are really, really hard to strip away.

45:34 Yeah, it is after all about people.

45:40 Years ago a politician got in trouble for basically saying corporations or people or

45:46 whatever the quote was that ultimately didn't land well for obvious reasons.

45:52 But companies aren't actually the topic that tends to matter.

45:59 It's the people within those companies where that pain manifests and if you can help solve

46:04 the problem for the people, then the companies will do business together because you are

46:10 actively making their life easier.

46:13 And that's what we strive to do and one of the things that just gets me the most excited

46:19 about my role today at Securoth is that the problem statement has evolved in identity

46:26 security.

46:27 It used to be just human actors and then you had workforce, workforce identity was one

46:33 problem statement.

46:34 And then it was customer identity where you're talking about basically people you pay money

46:37 to or people who pay money to you.

46:39 And the level of friction that they will tolerate and the level of access and credentials and

46:43 sort of customized user experience that you need to serve for them.

46:47 And now you have this new vector of delegated from any human type, delegated access, delegated

46:56 free will and an actor that has to be governed and attributable in some way.

47:01 It's a new problem, it's a new challenge and it's going to show up in different ways across

47:05 different verticals, across different organizations and it's really fun to be in a position that

47:12 can with confidence give guardrails to the whole problem statement.

47:17 And I think that's where legacy identity vendors are just missing the boat.

47:22 It's not like a startup that's just focusing on non-human identities can solve the problem

47:28 in full if they don't have credibility and confidence around all the human identities.

47:33 You have to be able to solve the whole picture and do so in a way that removes the pain and

47:39 allows the CISO to move from the department of no or this big red stop sign or whatever

47:46 it is, roadblock, to enthusiastically green lighting.

47:51 The adoption of these tools because there's confidence in the overall posture and the

47:55 new problem that has shown up in the last really two years.

48:00 Well, Mark, I really appreciate what you're doing in the space.

48:05 I think this is a whole new world for so many people.

48:08 And I think we haven't had, we're starting to, we haven't had too many big headline issues

48:14 just yet, but I know they're on the way and everything you're doing at SecureAuth I think

48:19 is going to give people the right guardrails to be productive, but keep their companies

48:23 and people safe as well.

48:25 And I really appreciate everything you shared on the go-to-market strategies.

48:28 There really aren't any hacks anymore.

48:30 I think you're not going to go crank out the email and LinkedIn spam cannons and get in

48:35 front of people and nobody's going to be responding to it.

48:37 It's the good old fashioned, put out content that adds value and builds trust.

48:43 Look for those warm connections, get customer case studies, get customers talking about

48:47 your product and show up in ways that have an organic connection.

48:52 And as always, I'm a firm believer sales isn't going anywhere.

48:55 I think there's a huge difference between buying from a bot and buying from a human,

49:01 especially if you're selling something that is mission critical to that person's job and

49:06 livelihood.

49:07 It's very tough to build that trust without a human interaction.

49:11 So Mark, thank you for being on the podcast.

49:14 Thank you for doing what you're doing.

49:16 Thank you for sharing what you shared and I can't wait to see what you and SecureAuth

49:20 do next.

49:22 Thank you very much.

49:23 It's been an absolute pleasure and it's really fun to speak to somebody who's like-minded

49:28 around the way we approach the industry and it'll be a lot of fun to go through this next

49:33 macro change with everybody as we all try to adopt AI and get confident in mitigating

49:41 the downsides of a new and exciting technology.

49:44 Yeah.

49:45 Well, things are changing so fast that we need to keep each other up to speed.

49:50 So I appreciate you sharing and thank you.

49:53 Great.

49:54 Thank you.