---
title: "The Token Bill Came Due: What Comes After the AI Pilot Phase"
episode: 132
podcast: "The LeanScale Podcast"
publisher: "LeanScale"
guest: "Anish Chadalavada"
guest_title: "Co-Founder & Chief Growth Officer, Gradial"
date_published: 2026-10-05
date_modified: 2026-10-21
duration: 00:42:31
word_count: 7389
topics: ["ai-in-gtm", "revenue-operations", "gtm-strategy", "brand-positioning"]
canonical_url: https://www.leanscale.team/knowledge/podcast/anish-chadalavada-gradial-the-token-bill-came-due/
source: "LeanScale Knowledge Hub — https://www.leanscale.team/knowledge"
license: "Free to quote and cite with attribution to The LeanScale Podcast."
---

# The Token Bill Came Due: What Comes After the AI Pilot Phase

_Anish Chadalavada on why phase one of agentic AI ended with a token bill nobody could explain, what an enterprise AI harness is, model lock-in as the new multi-cloud problem, multiplayer governance, which martech survives, and why the premium is moving back to humans_

**Episode 132 · The LeanScale Podcast**  
Anish Chadalavada, Co-Founder & Chief Growth Officer, Gradial · Hosted by Anthony Enrico  
Published October 5, 2026 · Updated October 21, 2026 · 00:42:31  
Canonical: https://www.leanscale.team/knowledge/podcast/anish-chadalavada-gradial-the-token-bill-came-due/

**Topics:** AI in GTM · Revenue Operations · GTM Strategy · Brand & Positioning


## Executive summary

Anish Chadalavada is co-founder and Chief Growth Officer at Gradial, which builds what it calls an enterprise AI harness: a layer that connects frontier models to the brand context, approvals, guardrails and tools a large marketing organization already runs on. Before Gradial he invested in early-stage AI companies at Point72 Ventures through the ChatGPT inflection and worked on AI strategy at Microsoft. According to the show notes, Gradial has since raised a $65M Series C led by Insight Partners and counts large enterprises such as AWS, T-Mobile, Prudential and Vanguard among its customers.

His starting claim is that the marketing SaaS stack won't survive the shift to agents, though not in the way the 'SaaSpocalypse' argument puts it. Marketing has always run on software that stores data but doesn't do the work, which has turned the job into project management, handoffs and tickets. Core systems of record that hold real context, such as the CRM, become infrastructure that agents draw on. Point tools that exist only to make another tool easier for a human to use lose their reason to exist, because software will mostly be used by agents.

The episode's title comes from Anish's view of where enterprises are now. The last twelve months were phase one: everyone tried pilots, tools, models and home-built agents. Then the quarter closed, the token bill arrived, and nobody could say what it produced. Content quality slid into slop that hurts the brand. Phase two is a control tower for agents, focused on a few critical 'hero workflows' and built top-down on context, guardrails, brand and process. He compares it to onboarding a new employee with a cheat sheet of the 20 things that matter.

Anthony Enrico brings his own version of the problem. LeanScale, about 30 people, had everyone on Claude Max plans. Some people ran out and moved to usage billing, and when he looked into why, it came down to infrastructure: indexing the whole CRM on every query, or running the latest model on simple tasks. Anish's answer is that single-player AI economics don't add up at the function level. Ten people building their own content-review agent duplicates token spend and breaks brand guardrails. Governance has the same root cause: most LLM products are designed for one user, so permissions have to be designed from the ground up for multiplayer use. He also argues model lock-in is the multi-cloud story again, since the best model for a task keeps changing.

The second half turns to people. Gradial connected its RevOps lead directly to business development, so the person who knows where leads are stalling decides who gets outreach and with what message. Anish sees consolidation of roles rather than deletion, and recommends generalist GTM roles or full-stack product-builder roles for people starting out. He explains why Gradial chose workflow over content generation, drawing on what he saw as an investor: founders solving hard problems outperformed those chasing easy ones. He closes with a contrarian prediction. As people tire of AI, they will pay a premium for human interaction, and AI's job becomes the back-office work that lets one advisor, marketer or nurse serve far more people.


## Key takeaways

1. **The marketing SaaS stack resets into infrastructure; it doesn't die** — Anish rejects the black-and-white 'SaaS is dead' view. What he expects is consolidation: point solutions that do a small piece of a campaign or make another product easier to use go away, while core systems of record that hold context become infrastructure agents use to do the work.
   _Why it matters:_ When reviewing a stack, separate the systems that hold real context from the tools that exist to manage other tools. The first group gains value as agents arrive. The second group is where consolidation will happen.
   _For:_ Marketing Leaders, RevOps Leaders, Revenue Executives

2. **Phase one ended when the token bill arrived** — Over the past twelve months enterprises tried pilots, products, models and home-built agents. At quarter end they got a large token bill and couldn't articulate what it produced. Agents became so easy to build that everyone built one, and content quality slid into slop that damages the brand.
   _Why it matters:_ Token spend that can't be traced to work outputs will be hard to justify. Anish expects companies to move from 'token maxing' to tracking AI spend against real artifacts and outcomes.
   _For:_ Revenue Executives, Founders, RevOps Leaders

3. **Phase two is a control tower for a few hero workflows** — Most customers are in the early days of phase two: a command center for agents that executes the critical 'hero workflows'. The approach is top-down rather than building agents as fast as possible. It starts with process and brand context (approvals, guardrails, brand) and with data context (content, its performance, audiences, products).
   _Why it matters:_ Pick the handful of workflows that slow the team down most and build the context around them first. Putting those two kinds of context together systematically is what Anish calls the fertile soil for agents.
   _For:_ Marketing Leaders, RevOps Leaders, Revenue Executives

4. **An AI harness connects the best models to the enterprise's context and tools** — Anish defines a harness as the layer that connects the best LLMs with the process, brand context, guardrails, tools and integrations an enterprise already uses, so agents can execute the workflows teams spend time on today, such as launching a back-to-school promotional campaign.
   _Why it matters:_ The aim is to give marketers time back for creative and strategic work. People stay at the steering wheel, because without them you get high volume and slop.
   _For:_ Marketing Leaders, RevOps Leaders

5. **Model lock-in is the multi-cloud problem again** — Large companies don't want to be locked into one cloud, and the same applies to model providers. Anish points to how quickly usage shifted from Claude toward OpenAI after a new OpenAI model release, visible in OpenRouter call data, and notes that Gemini may be best for image creation. A team that built everything around one provider has put itself in a risky position.
   _Why it matters:_ Use different models for different tasks and keep the ability to switch. That is one reason he argues against treating this as a pure build project.
   _For:_ RevOps Leaders, Founders, Revenue Executives

6. **LLM providers are paid on usage; enterprises are paid on outcomes** — Anish sees a basic incentive gap. Model providers want to drive usage. Enterprises want outcomes they can track. Spending heavily in a general AI product may make an individual more productive, but it doesn't show how that maps to what the business needs to solve.
   _Why it matters:_ This is where he sees the case for vertical applications: tracking work back to outputs and artifacts in a multiplayer way, with spend governance built in.
   _For:_ Revenue Executives, Founders

7. **Single-player AI economics break down at the function level** — Anthony describes LeanScale's roughly 30 people on Claude Max plans, some running out and moving to usage billing because they indexed the whole CRM on every query or used the latest model for simple tasks. Anish adds that when everyone in a function builds their own agent for the same problem, such as content review, the company duplicates work and token spend and weakens brand guardrails.
   _Why it matters:_ Solve shared problems once, centrally, at the function level, and let people keep using AI for personal productivity. Nobody should build the same thing ten times.
   _For:_ RevOps Leaders, Revenue Executives, Founders

8. **Governance has to be designed for multiplayer from day one** — Teams connect all their tools and then can't reconcile who has access to what. Most LLM products are productivity apps designed around a single user, which makes permissions hard. Anish says that whether you build, buy or use a vendor, you need a hierarchy from the start: shared organizational context everyone can see, layers of permission within it, and individual data that is not shared.
   _Why it matters:_ Mirror the permissions people already have in how they use AI. Retrofitting access control onto single-player tools is where data ends up shared in ways it shouldn't be.
   _For:_ RevOps Leaders, Revenue Executives

9. **The survival test for martech is whether it holds critical context** — Without naming vendors, Anish's barometer is that systems storing a lot of critical business context, such as the core CRM, have staying power because context is the fuel for agents. Tools that exist because configuration is hard, such as email builders layered on top of other systems, have less, because agents can use the original software directly.
   _Why it matters:_ The bloated stack came from adding a tool, then another, then an agency to make hard software usable. As agents take that on, agency and team hours can move to strategic work.
   _For:_ Marketing Leaders, RevOps Leaders

10. **Gradial put RevOps in charge of business development** — Anthony sees RevOps taking the lead on AI infrastructure in go-to-market, more than engineering or product. Anish agrees and goes further: Gradial connected its RevOps lead directly with the BDR team. When leads aren't getting outreach, the RevOps lead decides who to reach out to and with what message, because that person knows best what needs to happen for leads to become opportunities.
   _Why it matters:_ He describes this as consolidation of roles, not deletion. AI lets one person span insights, reporting, business development and some content, and operational roles such as RevOps and marketing ops can become strategic ones.
   _For:_ RevOps Leaders, Sales Leaders, Revenue Executives

11. **Entry-level paths: generalist GTM roles and full-stack product builders** — For technical graduates, Anish recommends a full-stack product-builder role that combines product strategy, customer conversations, design and writing code. For non-technical graduates, he suggests a flexible go-to-market role, something like GTM engineering, aimed at a concrete problem such as not creating opportunities or meetings fast enough.
   _Why it matters:_ What these roles require is critical thinking, strategic reasoning and scrappiness: test an idea, come back with results, iterate. Problem solving is the cross-functional skill he expects to keep its value.
   _For:_ Revenue Executives, Founders, RevOps Leaders

12. **Using AI doesn't remove accountability** — Anish describes the 'slop grenade': using AI without guardrails pushes the review and standard-setting burden onto whoever is downstream. Scaled across an organization, that taxes everyone's time and produces content that hurts the brand. Anthony adds a line from a former CRO, Tom Miller: practicing a bad golf swing more makes you worse than not practicing at all.
   _Why it matters:_ Let people use AI for productivity, but make clear that the person who ships the work is still accountable for it. Otherwise a company can end up worse off than it was 24 months ago, with a brand nobody associates with taste.
   _For:_ Marketing Leaders, Revenue Executives, Sales Leaders

13. **Solve the hard problem: why Gradial chose workflow over content** — As an investor, Anish saw founders fall for shiny-object AI use cases. Those solving easy problems got early traction but struggled to scale as the market saturated, while founders tackling hard problems succeeded at a higher rate than expected. When Gradial started, the focus in the market was content, image and video generation. Gradial chose process and workflow instead: the approvals, checks, dependencies, teams and agencies needed to get something simple out the door.
   _Why it matters:_ His argument is to remove the work people don't want to do, not the creative work they do want to do: 'Why would you want to eliminate content from marketing?'
   _For:_ Founders, Marketing Leaders

14. **The contrarian prediction: a premium on human interaction** — Rather than a world of personal agents talking to brand agents, Anish expects people to grow more fatigued with AI and to pay a premium to work with a human on things that matter, such as shopping, travel or financial advice. Anthony compares it to food: after ultra-processed convenience came a willingness to pay double for organic.
   _Why it matters:_ The B2B opportunity moves to automating the logistical and operational work behind a financial advisor, travel advisor, nurse or marketer, so a person who served 10 clients could serve 500 while staying the relationship holder.
   _For:_ Founders, Revenue Executives, Marketing Leaders

15. **Every function will get its own harness** — LLMs are horizontal and optimized for single-player productivity. Anish sees room for function-specific products in legal, support, sales and marketing that hold the context, workflows, approvals and guardrails of that function and orchestrate best-of-breed agents, models and people.
   _Why it matters:_ Because agents are easy to build and buy, the scarce piece is the control tower that knows the organization. Throwing more agents, people or tokens at a problem won't converge on the outcome by itself.
   _For:_ Founders, Revenue Executives, RevOps Leaders


## Frameworks

### SaaS Resets Into Infrastructure (01:45)

**Definition:** In an agentic stack, systems of record that hold critical context become infrastructure that agents use, and tools that exist only to make other tools easier for humans to use disappear.

Anish's alternative to the 'SaaSpocalypse' view. Marketing software has stored data without doing the work, and a web of point solutions grew to make hard software usable. Once agents can use the underlying systems directly, those usability layers have no job, while context-rich systems such as the CRM become more important because context is what agents run on.

### Phase One to Phase Two of Agentic Deployment (04:31)

**Definition:** Phase one was broad experimentation with pilots, tools, models and home-built agents, ending in a token bill nobody could tie to output. Phase two is a governed control tower that executes a small set of hero workflows.

Phase two is top-down. It starts by deciding what to preserve about current ways of working (approvals, guardrails, brand) and what context agents need (content, its performance, audiences, new products). Putting those together systematically is what lets agents execute work, much like giving a new employee a cheat sheet of the 20 things that matter and letting them learn from trying.

### Enterprise AI Harness (06:28)

**Definition:** A layer that connects the best LLMs with an enterprise's process, brand context, guardrails, tools and integrations so agents can execute the workflows teams spend time on today.

The harness provides what single-player AI products lack: interoperability across systems such as the CRM, a CMS and a ticketing system from different vendors; the ability to use different models for different tasks; spend governance; and output tracked as artifacts in a multiplayer way. People stay at the steering wheel to keep quality up.

### Model Lock-In Is the Multi-Cloud Story (09:13)

**Definition:** Avoid committing to a single model provider for the same reason large companies avoid a single cloud: the best model for each task keeps changing.

Anish notes that after a new OpenAI model release, OpenRouter call data showed a fast shift from Claude to OpenAI, and that Gemini may be the best for image creation. A team that spent six to twelve months building only around one provider has put itself in a risky position.

### Multiplayer Governance Hierarchy (14:09)

**Definition:** Design AI access in three layers from day one: shared organizational context everyone can see, permissioned layers within it, and individual data that is not shared.

Most LLM products are built for one user, which makes it hard to reflect who is allowed to see what. Anish says that whether you build, buy or work with a vendor, AI use should mirror the permissions people already have. It can be a fairly simple hierarchy, but it has to be designed from first principles as multiplayer.

### The Context Test for Martech Survival (16:22)

**Definition:** A tool that stores critical business context has staying power. A tool that exists only to make another tool easier for a human to use does not.

Context is the fuel for agents, so core systems such as the CRM stay central. Email builders and tools that exist because configuration is hard lose their purpose once agents use the underlying software. The interoperability across the martech suite is where Gradial focuses.

### The Slop Grenade (27:54)

**Definition:** Using AI without guardrails pushes review and standard-setting onto whoever is downstream, creating a tax on the organization's time and on the brand.

Anish credits the idea to someone he doesn't name. His response is to let people use AI for productivity while making clear they remain accountable for the output. Without that, an organization can scale slop, exhaust people internally and externally, and lose the association with taste that he says is all a brand has.

### Solve the Hard Problem (31:01)

**Definition:** Founders who take on hard problems succeed at a higher rate than those chasing easy ones, because easy problems attract saturated competition.

From Anish's time investing at Point72 Ventures. He cites Sam Altman and Y Combinator making a similar argument. It shaped Gradial's choice of process and workflow (approvals, checks, dependencies, agencies) over content generation, which was where most attention went at the time.

### The Premium Moves Back to Humans (35:03)

**Definition:** As AI fatigue grows, people will pay more for human interaction on things that matter, and AI's role shifts to the operational work behind the human relationship.

Instead of agent-to-agent commerce, Anish expects demand to see a human for travel, shopping or financial advice. The B2B opportunity is automating the logistics behind those people so one advisor who served 10 clients can serve 500.


## Quotes

_Speakers inferred from an undiarized transcript — verify before attributing._

> "Marketing has always been built around software that stores data but doesn't actually do the work."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (01:08)

> "So I see this as not the end of SaaS, more like a reset that turns SaaS into infrastructure so agents can use it for context to actually go do work."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (02:52)

> "At the end of the quarter, you get this massive token bill. You can't articulate all the outputs that came out of the token bill."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (04:09)

> "It's so easy to build agents that everyone's built an agent and you're starting to see even the quality of content is turning into slop."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (04:19)

> "If I can put those two pieces together in a systematic way, then I've created the fertile soil for agents to go actually execute work."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (05:29)

> "This is the same as the multi-cloud story. If you're a big company, you don't want to be locked into one cloud."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (09:13)

> "It all came down to that infrastructure. They're indexing our entire CRM every single time they ask a particular query, or they're using the latest model on super, super simple tasks."
>
> — Anthony Enrico, The LeanScale Podcast Ep. 132 (10:56)

> "You really have to design from the ground up for multiplayer."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (14:56)

> "Those I think don't have as much staying power because software is predominantly going to be used by agents and not by people."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (17:19)

> "We're starting to see this consolidation of roles. We don't see a deletion of roles in go-to-market."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (19:43)

> "At the end of the day, even if you use AI, you're still the accountable person."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (28:25)

> "People don't associate your brand with taste anymore, and brand is the only thing that matters."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (29:04)

> "A CRO I used to work with, his name's Tom Miller, he used to say, practicing a bad golf swing more makes you worse than not practicing at all."
>
> — Anthony Enrico, The LeanScale Podcast Ep. 132 (29:20)

> "Instead we decided to focus on process and workflow, which didn't seem really appealing at the time for many folks."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (32:27)

> "Why would you want to eliminate content from marketing? Why not eliminate the things that people don't want to spend time on today, and then empower them to go focus on the things that matter."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (33:08)

> "People are increasingly fatigued with AI and increasingly seeking and placing a premium on human interactions."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (35:03)

> "Whereas previously I might have been able to support 10 clients, tomorrow I could support 500 because all of my back office and operational work is done by AI."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (36:44)

> "It's easy to build agents. It's easy to buy agents, which means that you need a control tower that holds all the context."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (40:32)

> "If you just throw agents at the problem or you just throw people at the problem or you just throw tokens, none of those is going to converge to the outcome that the organization needs."
>
> — Anish Chadalavada, The LeanScale Podcast Ep. 132 (41:01)


## Practical advice by role

### Revenue Executives

- Ask what last quarter's token spend actually produced. If nobody can tie it to outputs, treat that as the end of the experimentation phase.
- Pick a few hero workflows and fund a governed, central way to run them, rather than letting every team build its own agents.
- Avoid committing to a single model provider; keep the ability to route different tasks to different models.
- Make accountability explicit: whoever ships AI-assisted work owns its quality.
- Plan for consolidated, more strategic roles rather than headcount cuts, and expect people to be needed at the steering wheel.

### RevOps Leaders

- Audit where tokens go. Common waste includes re-indexing the whole CRM on every query and running the newest model on simple tasks.
- Solve shared problems such as content review once at the function level instead of letting ten people build ten agents.
- Design AI access as a hierarchy from the start: shared organizational context, permissioned layers, and private individual data.
- Treat context-rich systems of record as infrastructure for agents and question tools that only exist to make other tools usable.
- Consider taking on execution, as Gradial's RevOps lead does with business development: decide which stalled leads get outreach and with what message.

### Marketing Leaders

- Start agent work with context: approvals, guardrails and brand on one side, and content, performance, audiences and products on the other.
- Keep people in the loop on anything that reaches the market. Scaled slop costs brand and taste.
- Point AI at the process work around campaigns (approvals, checks, dependencies, handoffs) rather than at removing the creative work.
- Review point tools and agency hours that exist only to make hard software usable; those are where consolidation will come from.

### Founders

- Choose a hard problem over an easy, crowded one. As an investor, Anish saw founders on hard problems succeed at a higher rate than expected.
- Expect function-specific harnesses (marketing, sales, support, legal) to be a durable category, because general LLMs are built for single-player productivity.
- Hire early-career people into generalist GTM or full-stack product-builder roles and judge them on problem solving and scrappiness.
- Bet on human relationships as the premium layer and use AI for the operational work behind them.


## AI takeaways

**Thesis:** Phase one of enterprise AI ended with token bills nobody could tie to output and a flood of slop. Phase two is governed: a function-specific harness that holds context, approvals and guardrails, routes work across models, tracks output in a multiplayer way, and keeps accountable people at the wheel.

- **** — 
- **** — 
- **** — 
- **** — 
- **** — 
- **** — 

**Agent & automation ideas**

- A central content-review agent for the whole marketing function that enforces brand guardrails, replacing individually built versions.
- A token-spend monitor that flags users who re-index the whole CRM on every query or use frontier models on simple tasks.
- A routing layer that sends each task to the model best suited to it, such as image creation to one provider and reasoning to another.
- A campaign-launch workflow agent that runs the approvals, checks and handoffs for something like a back-to-school promotion across CMS, CRM and ticketing.
- A stalled-lead agent that surfaces deals and leads without outreach for a RevOps or BDR owner, with a suggested message.


## Operations takeaways

### Revenue operations

- **.** 
- **.** 
- **.** 
- **.** 
- **.** 

### Pipeline & marketing ops

- **.** 
- **.** 
- **.** 


## Metrics mentioned

| Value | Metric | Context |
| --- | --- | --- |
| About 12 months | Length of the experimentation phase | The period in which enterprises tried many AI pilots, tools, models and home-built agents before token bills forced the question of what it produced. |
| About 30 people | LeanScale headcount on Claude Max plans | Anthony's own example: some people ran out and moved to usage billing because of how they used AI, not because of the work they were doing. |
| From 10 to 500 | Clients one advisor could serve | Anish's illustration of what happens when AI takes over back-office and operational work and the human stays the relationship holder. |
| Worse off than 24 months ago | How far back scaled slop could set a company | Anish's warning: scaled AI slop exhausts people internally and externally and erodes the brand's association with taste. |
| $65M, led by Insight Partners | Gradial Series C | From the episode's show notes; Anthony congratulates Anish on the Series C at the close. |
| 80–90% | Campaign execution time reduction | From the show notes: what one of Gradial's enterprise customers achieved. Not discussed in the conversation itself. |


## Entities mentioned

- **Gradial** (company) — Anish's company, where he is co-founder and Chief Growth Officer. Gradial builds an enterprise AI harness for marketing that connects frontier models with brand context, approvals, guardrails and the tools an enterprise already owns. It chose process and workflow over content generation, focuses on the enterprise because governance is hardest there, and connected its own RevOps lead directly to its BDR team. Anthony congratulates Anish on the Series C at the close. · https://www.leanscale.team/knowledge/company/gradial/
- **Point72 Ventures** (company) — Where Anish invested in early-stage AI companies around the launch of ChatGPT. Watching founders there, he concluded that those taking on hard problems outperformed those chasing easy ones, which shaped Gradial's focus on workflow. · https://www.leanscale.team/knowledge/company/point72-ventures/
- **Updata Partners** (company) — Anthony mentions a recent CEO summit with Updata Partners, where the shared view was that most companies' AI strategy is really a data model problem. · https://www.leanscale.team/knowledge/company/updata-partners/
- **OpenAI** (company) — Cited by Anish in the model lock-in argument: after a new OpenAI model release, usage shifted quickly from Claude to OpenAI. He also uses it as the example of an LLM provider incentivized to drive usage. · https://www.leanscale.team/knowledge/company/openai/
- **Anthropic** (company) — Maker of Claude, the model many teams had built around over the previous six months, and of the Claude Max plans LeanScale had its people on. · https://www.leanscale.team/knowledge/company/anthropic/
- **Y Combinator** (company) — Anish cites Sam Altman and Y Combinator as having argued that solving a hard problem is better for a company than solving an easy one. · https://www.leanscale.team/knowledge/company/y-combinator/
- **LeanScale** (company) — Anthony's company, about 30 people. Everyone was on Claude Max plans, and some ran out because of how they used AI: indexing the whole CRM on every query or using the latest model for simple tasks. · https://www.leanscale.team/knowledge/company/leanscale/
- **Anish Chadalavada** (person, guest) —  · https://www.leanscale.team/knowledge/guest/anish-chadalavada/
- **Anthony Enrico** (person, host) — Co-founder of LeanScale and host of The LeanScale Podcast. · https://www.leanscale.team/knowledge/guest/anthony-enrico/
- **Tom Miller** (person, mentioned) — 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. · https://www.leanscale.team/knowledge/guest/tom-miller/
- **Claude** (tool, AI Assistant) — LeanScale had everyone on Claude Max plans and some people ran out. Anish notes many teams built around Claude over the prior six months before usage shifted toward OpenAI.
- **ChatGPT** (tool, AI Assistant) — Anish's marker for the AI inflection: he was investing at Point72 Ventures in the months before and after ChatGPT came out.
- **Gemini** (tool, AI Assistant) — Given as an example of a model that may be best for a specific task, image creation, which is why Anish argues for using different models for different tasks.
- **OpenRouter** (tool, AI Infrastructure / Model Routing) — Anish points to OpenRouter model-call data as showing how quickly usage shifted from Claude to OpenAI after a new model release.
- **GitHub** (tool, Developer / Code Hosting) — Anthony mentions teams that think their data is organized because they put everything into a shared GitHub, as an example of infrastructure that may not be enough.
- **LinkedIn** (tool, Social Platform) — Anthony says he loves LinkedIn's 'This is AI' button; both cite LinkedIn and email inboxes as where AI slop content is most visible.


## FAQ

**Q: What is an enterprise AI harness?**

A: An enterprise AI harness, as Gradial co-founder Anish Chadalavada describes it, is a layer that connects the best large language models with a company's process, brand context, guardrails, tools and integrations so that agents can execute the workflows teams currently spend time on, such as launching a promotional campaign. It handles what single-user AI products don't: interoperability across systems like the CRM, CMS and ticketing, using different models for different tasks, spend governance, permissions, and tracking output as shared work artifacts. People remain at the steering wheel to keep quality high.

**Q: Will AI agents kill marketing SaaS?**

A: Not entirely, according to Anish Chadalavada of Gradial. He sees a reset rather than an end. Core systems of record that hold critical context, such as the CRM, become infrastructure that agents rely on, because context is the fuel for agents. Point solutions that do a small piece of a campaign or exist only to make another tool easier for a human to use, such as some email builders, lose their purpose because software will mostly be used by agents rather than people.

**Q: Why are companies struggling to justify AI token spend?**

A: After about twelve months of experimenting with AI pilots, tools, models and home-built agents, many companies received large token bills at quarter end without being able to say what the spend produced. Anish Chadalavada says LLM providers are incentivized to drive usage, while enterprises need outcomes they can trace. Spend becomes hard to justify unless it is tied to real work outputs in a shared, governed way. Common waste includes everyone building their own version of the same agent and, as LeanScale found, re-indexing the whole CRM on every query or using the newest model for simple tasks.

**Q: How should companies govern AI access to sensitive data?**

A: Design for multiple users from the start, says Anish Chadalavada. Most LLM products are built as single-user productivity apps, which makes it hard to control who sees what once tools are connected. Whether you build, buy or use a vendor, set up a hierarchy that mirrors existing permissions: shared organizational context everyone can access, permissioned layers within it, and individual data that is not shared.

**Q: Should a company build everything on one AI model provider?**

A: Anish Chadalavada compares this to multi-cloud: large companies avoid locking into one cloud, and the same logic applies to models. The leading model changes quickly. He notes usage shifted rapidly from Claude to OpenAI after a new OpenAI release, and that a different provider may be best for tasks like image creation. Teams that built only around one provider are exposed, so the ability to use different models for different tasks is a core part of an AI harness.

**Q: Who should own AI infrastructure in a go-to-market team?**

A: Increasingly RevOps. LeanScale co-founder Anthony Enrico sees RevOps taking the lead more often than engineering or product, and Anish Chadalavada agrees because RevOps manages the data and context of the revenue lifecycle. At Gradial, the RevOps lead also works directly with the BDR team and decides which stalled deals and untouched leads get outreach and with what message, an example of roles consolidating rather than being eliminated.

**Q: What entry-level jobs make sense in go-to-market as AI spreads?**

A: Anish Chadalavada recommends a generalist go-to-market role for non-technical graduates, something like GTM engineering, aimed at a concrete problem such as not creating opportunities or meetings fast enough, where the tools to test ideas are now widely available. For technical graduates he suggests a full-stack product-builder role that combines product strategy, customer conversations, design and code. The common requirements are critical thinking, strategic reasoning and scrappiness.


## Timeline

- **00:00** — Cold open and intro
- **00:58** — The claim: the marketing SaaS stack won't survive agents
- **03:05** — Anthony's pushback: it's a data model problem, not a model problem
- **03:45** — Phase one is over: the token bill came due
- **06:14** — What an enterprise AI harness actually is
- **08:12** — Spend governance, and model lock-in as the multi-cloud story again
- **10:38** — Why single-player AI economics don't add up
- **13:16** — Governance and multiplayer architecture
- **15:56** — Which martech survives agents
- **18:26** — Who owns AI inside go-to-market
- **19:10** — Gradial put RevOps in the driver's seat of business development
- **22:37** — Entry-level GTM roles in an AI world
- **27:39** — The slop grenade and accountability
- **29:52** — From Point72 Ventures to founding Gradial
- **32:14** — Why Gradial chose workflow over content generation
- **34:28** — What happens next: the premium moves back to humans
- **38:59** — A harness for every function
- **41:12** — Wrap


## Related episodes

- **Ep. 126: The Agents Are the Easy Part** (Bastien Botella (Fountain)) — Another view of agentic marketing in production, where the hard part is the data and context layer and model choice is sized to cost, matching Anish's harness and token-bill argument.
- **Ep. 122: Your AI Strategy Is a Data Model Problem** (Jake Toepel (LeanScale)) — Anthony opens with this framing, that AI strategy is really a data model problem, before Anish argues context is the fuel for agents.
- **Ep. 119: The Agent Was Never the Hard Part** (Kushal Sharma (Circle)) — Builds on the same point Anish makes: agents are easy to build, and what decides success is the context and governance around them.
- **Ep. 111: Your AI Agents Have God Mode (And Nobody Vetted Them)** (Mark van Oppen (SecureAuth)) — A deeper look at agent access and governance, the multiplayer permissions problem Anish says has to be designed in from day one.
- **Ep. 105: You Built a Brilliant AI Agent — Now Get 300 People to Use It** (Christopher O'Donnell (Day AI)) — Moving AI from individual productivity to organization-wide use, the single-player versus multiplayer shift Anish describes.
- **Ep. 123: Stop Hiring GTM Engineers. They Already Work For You.** (Ryan Vanshur (Handle)) — Consolidating GTM roles and giving operators more execution responsibility, as Gradial did by putting RevOps in charge of business development.
- **Ep. 53: AI vs. Human Connection: The Future of Sales & Marketing** (Khurram Kalimi (VinnCorp)) — Another take on Anish's closing prediction that AI fatigue will put a premium on human interaction. · https://www.leanscale.team/knowledge/podcast/khurram-kalimi-ai-human-connection/
- **Ep. 100: AI Ops: How We Run RevOps for 30 SaaS Companies at Once** (Jake Toepel (LeanScale)) — LeanScale's own approach to running AI centrally at the function level, which connects to Anthony's Claude Max token story in this episode.


## Full transcript

_Machine-transcribed and not diarized; speaker attribution is inferred._  
_Transcript only, as a separate file: https://www.leanscale.team/knowledge/podcast/anish-chadalavada-gradial-the-token-bill-came-due/transcript.md_

### 00:00 — Cold open and intro

**[0:00]** My guest today is Anish Chhatalavada, Co-Founder and Chief Growth Officer at Gradial, the company building the enterprise AI harness that runs marketing work across all the tools an enterprise already owns.

**[0:13]** See this as not the end of SaaS, more like a reset that turns SaaS into infrastructure.

**[0:30]** It's really just the unit economics fundamentally don't make a lot of sense when it comes to purely thinking of it as a single player operation.

**[0:41]** Whereas if you just throw agents at the problem or you just throw people at the problem or you just throw tokens, like none of those is going to converge to the outcome that the organization needs.

### 00:58 — The claim: the marketing SaaS stack won't survive agents

**[0:58]** Anish, you said the traditional marketing SaaS stack won't survive the shift to AI agents. That's a pretty big claim. What makes you so convinced?

**[1:08]** Yeah, that's a good question, Anthony. So I'd say marketing has always been built around software that stores data but doesn't actually do the work.

**[1:18]** And over time we start seeing this create a bigger and bigger tax on marketers and marketing has become more about project management, handoffs, tickets, tools and tracking than it has about actually ideas and building great customer experiences.

**[1:37]** I think there's been this kind of black and white perspective in the market around the SaaS apocalypse like, oh, SaaS is dead. No, it's critical.

**[1:45]** And really when I talk about the marketing stack and how agentic is going to disrupt that, really what I'm referencing is a simplification or consolidation where all of these point solutions and tools that exist to do a small piece of a campaign

**[2:05]** that make another product easier to use, those are going to go away. Your core systems of record that have context are essentially going to turn into infrastructure.

**[2:17]** And so when you think about the agentic marketing stack, you have something that's a lot simpler than the messy web most enterprises are accustomed to dealing with today.

**[2:28]** And so that's why SaaS apocalypse isn't quite correct. There will be software products that are holding critical context that are a key part of this story for sure.

**[2:37]** But all the point solutions that are required today to get everything to work that are just storing data or adding another checkpoint to this long list, those aren't going to be necessary anymore when you have agents that can go actually execute tasks.

**[2:52]** So I see this as not the end of SaaS, more like a reset that turns SaaS into infrastructure so agents can use it for context to actually go do work.

### 03:05 — Anthony's pushback: it's a data model problem, not a model problem

**[3:05]** I was actually recently just at a CEO summit with up data partners and we have a really similar point of view where I think most people's AI strategy has nothing to do with their ability to use the models, build agents, build workflows.

**[3:23]** It all is a data model problem. So, hey, do you have a semantic layer context graph data infrastructure even built in a way that AI can use? Where do you think people tend to fit kind of on the spectrum that you're talking about?

**[3:37]** Like, as you look at their stack, what do you think they have in place and what moves do you think they're going to need to make in the near future?

### 03:45 — Phase one is over: the token bill came due

**[3:45]** Yeah, I'd say everyone is still in the early days when it comes to getting real value from agentic. I think we're wrapping up the end of a long period of experimentation and trying a lot of different things because over the past 12 months, you know, everyone's tried a lot of different AI pilots, products, tools, different AI models, building things themselves.

**[4:09]** And then the token bills came to you and you start to see, you know, at the end of the quarter, you get this massive token bill. You can't articulate all the outputs that came out of the token bill.

**[4:19]** It's so easy to build agents that everyone's built an agent and you're starting to see even the quality of content is turning into slop. It's really detrimental for the brand.

**[4:31]** And so as we kind of reach the end of what I think of as phase one with these agentic deployments, most customers are in the early days of phase two, which is how do I actually have a control tower or command center for agents that allows me to go focus on executing work for those hero workflows or those critical workflows?

**[4:54]** And when you start thinking about the problem in that way, it's not so much how do I build agents as quickly as possible. It's a much more organized top down approach that boils down to, like you said, context, guardrails, a brand and process context.

**[5:12]** So how do I think about, you know, my current way of doing things and what I want to preserve about it, like the approvals, the guardrails, the brand, what's the context that the agents need about, you know, my content, the performance of that content, my audiences, how I think about the newest products.

**[5:29]** If I can put those two pieces together, and that's why we think of Gradial as a harness is if I can put those two pieces together in a systematic way, then I've created the fertile soil for agents to go actually execute work.

**[5:42]** It's very similar to onboarding a new employee. You know, if you throw a ton of different things at them, it's not possible for them to be successful. But if you give them a really good cheat sheet of here's the 20 critical things, and then I want you to try and then learn from what you've tried.

**[5:58]** Those, those people can onboard much more quickly and so I think of agentic as as essentially very similar to that process of how do I go get agents in an organized way, operating within the guardrails of the business to go execute work.

### 06:14 — What an enterprise AI harness actually is

**[6:14]** And you described this as a enterprise AI harness for, I think I have a rough idea of what that is, and what it looks like but what does that all entail.

**[6:28]** Yeah, so a harness is essentially connecting the best LMS with the process brand context guardrails and tools and integrations that an enterprise uses, so that you can actually go execute the workflows that their teams are spending time on today.

**[6:49]** And in doing so you can unlock time for those marketers to focus on the creative, the strategic the impact and like I said, there's a huge amount of demand and need for that.

**[7:02]** These marketing roles need people in them because without people there at the steering wheel you get a lot of slop and a lot of high volume of content but not necessarily like stuff that's a creative for the brand and so a harness is essentially putting all of the pieces

**[7:19]** of the puzzle together and allowing agents to work within that framework by connecting all of a brand's context and systems and guardrails with the LMS that can actually reason and plan, so that you end up with this agent that can actually go and execute workflows at scale,

**[7:37]** launching a promotional campaign for back to school, for example, for some of the teams that think they're running this natively out of cloud or open AI they think they have their data organized or, hey, I've put everything into a filtering system or shared GitHub.

**[7:57]** What are some of the things that people like, oh my gosh I didn't realize that I'm actually not able to do XYZ with this infrastructure, and we're going to need something that really reels in all of the AI horsepower that I'm driving.

### 08:12 — Spend governance, and model lock-in as the multi-cloud story again

**[8:12]** Yeah, I'd say it's the, the number one thing is, you know, you touched a little bit on the spend controls and the governance I think that there's like fundamentally this incentive model difference between LLM providers that are looking to drive usage right

**[8:29]** into the product, and then the actual desire to deliver enterprise outcomes and that's where the vertical applications, the need for vertical applications comes from is that, you know, if I spend a lot of tokens in open AI I am, I am definitely becoming more productive

**[8:46]** in my work but how do I track that back to the things that my enterprise actually needs to be able to solve for and so we're starting to move past this phase of, you know, token maxing is, is the best thing I can do because those tokens are expensive and if you can't

**[9:01]** track that back to actual work outputs in a systematic way and artifacts in a multiplayer way, then you're going to struggle to justify the ROI and struggle to justify that spend within the business.

**[9:13]** But the even more important part is just about that the interoperability like this is the same as the multi cloud story if you're a big company.

**[9:20]** You don't want to be locked into one cloud so a lot of times you know you have infrastructure and different clouds and model providers are the same like we had a lot of, you know, we heard a lot about the desire to build around clock over

**[9:33]** the past six months and then when when chat GPT six when Astra came out, then there was a big shift from, you know, you can see this in the open router model calls there was a big shift from Claude to open AI very very quickly so if I've built over

**[9:48]** the past six 12 months all around Claude, and then it turns out that you know open AI now has the most intelligent model and then Gemini is the best at image creation then I've indexed myself into this, this risky position that I shouldn't be in and

**[10:03]** so being able to use different models for different tasks is another critical component of a harness and a critical piece of why you don't want to see this as a purely build out endeavor because you want to be able to operate across these different

**[10:19]** models you want to be able to rely on vendors who really specialized in a particular functional area, and you want to be able to have those guardrails and spend governance built in, which is, is just challenging to do when you're driving

**[10:35]** consumption as quickly as possible.

### 10:38 — Why single-player AI economics don't add up

**[10:38]** I think it makes a ton of sense, and I am just starting to feel the pain. So lean scale or about 30 people. We had everybody on a bunch of Claude max plans, and what ended up happening is a lot of people would run out and then they're running

**[10:56]** into usage, and, and I'm when we go dig in like hey why are you using so many tokens, and then this person's barely using any but seems like you're getting the right job done. It all came down to that infrastructure like oh they're indexing our

**[11:11]** entire CRM every single time they ask a particular query or they're using like the latest model on super super simple tasks, and just like letting it run. And I think that's, we're still in this period if we're here in September, 2026 where a

**[11:29]** lot of this is getting subsidized. I'm under the impression that's going to change relatively quickly and the prices are going to go up soon.

**[11:38]** Yeah, definitely. It's really just the, the unit economics fundamentally don't make a lot of sense when it comes to purely thinking of it as a single player operation because really we see a lot of people that are like you said we building

**[11:55]** a wheel over and over again where you know a lot of times, and within an organization or at least within a function. There's common problems that everyone's dealing with, and you want to solve for them holistically.

**[12:08]** So if you have a situation where everyone has a cloud max plan and everyone's kind of got the same problems around content review for example, and they're all going and building their own agents and workflows for content review.

**[12:20]** One, you're starting to break down on the brand and guardrails you really want to centralize that and two, you're starting to duplicate work and duplicate token spend to accomplish the same thing and that that's really why you want to think about something

**[12:36]** holistically at the functional level of like hey I want to think about a harness for marketing or for sales or for support, where all of the hero workflows that I know are slowing my team down can be executed in a more central governed way, where I can track and control

**[12:53]** the end. And then, you know, folks can still use AI for their own productivity but it's not like everyone needs to go build the same thing over and over again, 10 times within within the work.

**[13:07]** Yeah, it makes a ton of sense and I, I had just seen so much recreation of the wheel over and over again, and it's so inefficient.

### 13:16 — Governance and multiplayer architecture

**[13:16]** Now, one of the things I'd love your perspective on too. It's not the sexiest topic, but governance. I mean I'm running into a lot of potential issues where, hey you're storing really sensitive data over here and then connecting it over there and then

**[13:33]** somebody can potentially get access to it. Or it's a little overly restrictive and certain things that people need to do they can't get access to the right things. What is, what is your thought on the governance infrastructure, and how do you give people the right access

**[13:49]** to the right data and workflows and agents, compared to what kind of seems to be like a free for all right now.

**[13:57]** Yeah, totally. I we see this all the time where it's like, everyone just goes and connects all their different tools and then it's hard to reconcile against like, who has access to what and then are people even putting data in the right places and this is this

**[14:09]** is part of the reason why, you know, really focused on the enterprise, because this is a really hard problem and a really hard pain point in the enterprise and really what we're looking for here is, how do I essentially reflect the permissions that people

**[14:26]** have and the rights that people have to see different information in the way that they use the AI products and that's fundamentally a multiplayer question. And because, you know, a lot of these LMS are architected for for driving usage and for simplicity

**[14:41]** and they're like these super productivity apps, they are really designed around single player and because they're really designed around single player it's just hard, it's just hard to solve that problem for, you know, who has access and like how do we make sure that data

**[14:56]** is not being shared in a way that it shouldn't, you really have to design from the ground up for multiplayer, whether you're building or buying or working with a vendor you really need to solve from day one for that multiplayer approach and say hey here's, here's the

**[15:11]** shared organizational context that everybody has access to. And then here's the different like layers of permission or access within that. And then here's the individual data that people have access to which should not be shared and, you know, it can be like a relatively

**[15:25]** simple hierarchy we benefit from, you know, we're working with with marketing teams and so a lot of times, there's content is intended to be public or is already public. And so, you know, we still see the same issues but you really do need to think of it from first principles

**[15:41]** as like a multiplayer architecture, in order for in order for that data and governance to be built in the correct way, bringing this back to marketing and specific for their use cases.

### 15:56 — Which martech survives agents

**[15:56]** You mentioned, hey, some of the tools will remain some of them won't remain. Are there any specific marketing, tech infrastructure platforms that you think will still see in the next two, five years.

**[16:13]** And are there any ones that you think like a something like this is probably not going to survive unless they do something dramatically different.

**[16:22]** Yeah, I'd say like the, the barometer for me on that without without naming names here is that if you're a, if you're a key system for the business that stores a lot of critical context.

**[16:34]** I think that there's a lot of staying power around that because we know context is the fuel for agents and without good context that's organized in a structured way that the business recognizes and understands.

**[16:46]** It's difficult for agents to be successful so when you think of their core CRM systems where all of this customer information lives. I think that those systems are really valuable and critical for agents to work correctly.

**[16:59]** And, you know, what we're solving for is that how do you go interoperate across like, you know, CRM, and then a different, you know, CMS from another brand and then a ticketing system that comes from another product provider that interoperability across that martech suite that glue that that's really challenging and that's where we've, we've really focused.

**[17:19]** If you're a system that purely exists to make another system easier to use for example, like there's, you know, the email builders or like, you know, tools that are existing because configurability is hard like those I think don't have as much staying power because software is predominantly going to be used by agents and not by people.

**[17:41]** And so once agents can use the software effectively tools that exists solely for human usability because the original software was too hard to use.

**[17:51]** That isn't going to be something that we see anymore but that that's the reason for like the bloated web of software most people are dealing with is, you know, I start with some software that it's too hard to use so I add another tool that makes it easier to use and then another tool and then an agency.

**[18:06]** So we're going to start to see, you know, I don't need to do all of those things I can focus my work, my agency work on the strategic impact that matters, and not putting hours of my team's time or my agency's time towards help make this system work for me.

### 18:26 — Who owns AI inside go-to-market

**[18:26]** Makes a ton of sense. From your perspective, who tends to be, because I feel like this is an interesting thing I'm seeing, I'll share what I'm seeing first and then would love your take.

**[18:37]** The ownership of AI infrastructure abilities that the team that's putting in the tools managing them in a lot of the companies, at least from a go to market perspective, I am seeing rev ops take a leadership role in there.

**[18:54]** I was anticipating it's not what I was expecting. I was expecting engineering or product teams or the CTO to kind of own that.

**[19:03]** Are you seeing the same thing for these go to market teams who tends to be picking up the ball on these AI projects.

### 19:10 — Gradial put RevOps in the driver's seat of business development

**[19:10]** Yeah, I'd say, I'd say rev ops is definitely like a very critical piece of the puzzle because they are managing a lot of the data and context for the business and like how, how the revenue lifecycle is moving and how the funnel of the pipeline is moving and you know we've

**[19:28]** actually taken an interesting approach of connecting our rev ops team directly with our business development and sales team so that when we see like deals are sold or leads are not getting outreach to our rev ops lead can then go work directly with the BDR team to actually

**[19:43]** go reach out to those leads. And so we've actually gone and empowered our rev ops team to take on more execution responsibility because I think over time, we're starting to see this consolidation of roles we don't see a deletion of roles and go to market I think there's

**[20:01]** a lot of anxiety around that but we don't we don't see a deletion of roles. Like I said if anything it's clear that people are needed more than ever based on some of the content that we've all seen in our email inbox and on LinkedIn right it's very

**[20:14]** detrimental for the brand for that content to go out at scale. So we actually need more people in the driver's seat than before. It's just that the time that they're focused, the things they're focused on can start to shift to higher value higher impact work.

**[20:28]** And we're also seeing in parallel this, this kind of consolidation in the sense of more generalist roles, where people are able, because of AI they're able to span across multiple functions or focus areas whereas before they were maybe rev ops is

**[20:44]** just rev ops today rev ops can be rev ops and all the insights and reporting and data and analysis, but also business development, and maybe even some content because that rev ops lead knows better than anyone, what needs to happen in the pipeline for deals to move

**[21:02]** or for leads to turn into opportunities. And so why not go empower that person to say you're responsible for for biz dev then you, you should say, who we should reach out to and with what message and that's working really well for us and it's a small

**[21:16]** example of like our view of the future state of the world for the product that we sell as well where it's like, maybe yesterday you were a content author who is responsible for building web pages. Well tomorrow, you have the most intimate understanding of the website so you should be the person who's maybe

**[21:31]** also responsible for not just the build with the help of agents but actually the content and the, the strategy for the website as well for both an audience of agents and an audience of people.

**[21:41]** And so, that used to be three or four different roles of project management content authoring content strategy, and then audience strategy and now it's like, let agents do the hard part of like the actual build and analysis and the manual work that's

**[21:57]** describable, and then you focus on the strategy across all of those. That's really what we're starting to see so in a nutshell to kind of go back to what you were talking about a second ago.

**[22:07]** I definitely see rev ops as one of the kind of key, key audiences for like the strategic like entry point for a business, but also other functions within go to market can start to see themselves and operate in a more strategic way, where you know previously maybe

**[22:23]** the operational role like marketing ops, similar to rev ops it's like well now you can be a very strategic role that controls your destiny versus being an operational center that is constantly being asked to do more with less.

### 22:37 — Entry-level GTM roles in an AI world

**[22:37]** Do you think I I'm, I'm seeing something really similar to consolidating into that general type of role also moving the strategic skill needed higher. So, we're needing to see those skills. One, one question that always makes me jump to is what are the good

**[22:57]** entry level positions for people who are just coming out of school, just getting into the workforce, typically you don't have a lot of experience to pull from to have a strong opinion on strategy.

**[23:09]** Are there things that are opening up for people in that position that you're seeing.

**[23:15]** Yeah, definitely I think a generalist go to market role is one of the best things, folks can do, as they come right out of college for example and they're, if they're not a technical person if you're a technical person, then I think being a product builder,

**[23:30]** is kind of my vision of like the full stack product role of, I'm going to do some product strategy and I'm going to talk to customers but I'm also going to do some design and wire framing on what this product experience should look like because that interface is actually a really critical differentiator

**[23:45]** of AI, and then I'm actually also going to write code as well, whether or not I studied computer science that full stack product builder role. If you're technical, I think is a really good place to start and you learn the strategy as you go, you know there's lots of room and opportunity

**[24:00]** to use scrappy and brings in a new way of doing things because they don't have the experience and they'll learn things on the job and I think something in parallel on the go to market side like I don't want to say necessarily like a GTM engineers the manifestation

**[24:15]** but something akin to that where hey I come in, you know I, I'm not a technical person per se but that doesn't matter for my ability to write code anymore so I can come in and basically solve a go to market problem because my problem growth is it, you know, something

**[24:31]** granular like, hey I can't create opportunities fast enough I can't create meetings fast enough top of funnel mid funnel pipeline like there's a kind of flexible go to market role where you can put someone in who, again, doesn't have experience but brings a new perspective

**[24:45]** and new way of doing things, and has all the tools at their disposal that they can go, go figure it out. And if you want to test some ideas it's easier to do that now than it ever has been in the past.

**[24:56]** So go test something out and then come back with the results and really seeing that as a full stack, a full stack role and all it requires is, you know, the ability to reason and think critically and think strategically and, and really just be scrappy and use all the tools

**[25:13]** at your disposal to go solve problems, think problem solving is that ultimate, you know, ultimate cross functional domain that there, there will be space for you, even with all the progress with AI like someone who can think about solving problems

**[25:28]** be scrappy and go test ideas and then go iterate on those ideas, there's, there's plenty of room and opportunity for those types of roles and it's a good fit for folks who are early junior career, maybe just starting out in their careers.

**[25:43]** It gives me a lot of hope I have three kids another one on the way in November, and I'm just thinking like, what are their opportunities going to look like I know it's not going to look what it is today. I just hope there's something that they can aim towards.

**[25:57]** 100% I definitely think there is and like I said I see a lot more. I see more and more room and opportunity for people to do the roles like that, that we see as as being critical today, where it's just the, the role is evolving, right, it's just evolving

**[26:16]** because we don't need as much human time spent on lower impact tasks that are repetitive, which is good. I think that was a lot of the promise of the internet was to do that, which, you know, it's dubious like how much productivity was gained from the onset of the internet

**[26:34]** right like now there's a lot of research pointing to this idea that there's some productivity gains but a lot of cases, the work just shifted medium from like pen and paper or from Excel to like a software application but people were still doing repetitive row tasks over

**[26:52]** again there's still people as of, you know, a couple years ago who are still doing data entry but they were just instead of doing it in a database they were doing it in like a UI. And so now with AI it's like AI can actually do the function of data entry.

**[27:06]** But that doesn't mean that there's not people who should be looking at data goes back to the rev ops and marketing ops question of like, we still want people looking at the insights and coming up with the strategy of like, how should I operate my business, what's the best way to connect

**[27:20]** people. And that's not something that AI has eaten up or is about to eat up, in my opinion, if anything, we've gotten to a point of fatigue with AI and people are starting to like, you know, band slop content right or automatically market a spam and so we actually

**[27:37]** really need people there.

### 27:39 — The slop grenade and accountability

**[27:39]** Yeah, I have definitely seen that I love the LinkedIn. This is AI button. It's one of my favorites. Yeah, yep, it's everywhere now and I love it too. I think it's, I think it's great and I think it's like, you know, this, I forget who talked about this idea of like

**[27:54]** the slop grenade which is like, you know, the use of AI creates the use of AI on its own right creates attacks on people downstream because it's like I'm offloading the review and the bar setting that the standard setting I'm offloading that from myself to someone downstream

**[28:13]** or someone who reviews my work, if I just use AI to do it. And so, how do we then go and allow people to use AI for productivity but then empower them to say no, you're still the accountable person.

**[28:25]** At the end of the day, even if you use AI you're still the accountable person.

**[28:29]** So we really do need that mechanism of action in the enterprise and just everywhere because otherwise, you know, we risk this like broad tax that happens on on society and on teams where the use of AI by you know the people doing the work without guardrails

**[28:47]** and strategy, then creates attacks that eats up the entire organization's time, and then you end up with scale slop content that actually hurts your brand. And so at the end of the day you're actually worse off than you were 24 months ago because you've put out all

**[29:04]** the slop content you've fatigued and exhausted everyone internally and everyone externally, and people don't associate your brand with taste anymore and brand is the only thing that matters. So, it's a big risk.

**[29:20]** A CRO I used to work with, his name's Tom Miller, he used to say, "Practicing a bad golf swing more makes you worse than not practicing at all." And I always think about that when, like you mentioned, "Hey, if you're cranking a bunch of slop out into the market, it's incrementally making your brand worse."

**[29:42]** So, you'd be better off just doing nothing and just then continuing to throw a bunch of slop into the market.

### 29:52 — From Point72 Ventures to founding Gradial

**[29:52]** Exactly. I'd love to hear the story because you have a very decorated career, a long time at point 72, I think going from that investor position into an operator role. I'd love to hear how that happened and what were you seeing that made you think that this was such a big opportunity?

**[30:14]** Yeah, I was investing in early stage AI companies and it was a really exciting time to be working on AI. It was right at the cusp of chat GPT and the kind of six months proceeding and immediately after chat GPT came out.

**[30:29]** And so it was clear that there was this new paradigm shifting technology that was available, but a lot of the use cases and applications still wasn't clear. It also wasn't clear the pace at which AI would improve and what would become possible in what time period.

**[30:45]** But it was clear that there was a transformative new technology and there was an opportunity that hadn't existed before. And so that kind of got me thinking about starting a company and thinking about what might be some interesting problems to go work on.

**[31:01]** One thing I saw a lot of early stage founders start to a little bit fall victim to is the shiny object syndrome a little bit of, "Hey, I want to go just work on AI and I'm going to go solve for a pretty easy use case."

**[31:18]** I think, you know, Sam Altman and the Y Combinator folks have talked a lot about how solving a hard problem is actually better as a company than trying to go solve an easy problem because there's so much competition around the easy problems.

**[31:34]** And there's more room to explore, fail, iterate, and become an expert when you're going and solving a hard problem than it is when you're going and solving an easy problem. And so paradoxically, not that building a company is ever easy, but paradoxically, you should as a founder go solve something that's challenging and/or is close to home.

**[31:57]** And I think I really internalized that from seeing the companies that were successful is that they were going and solving a hard problem in most cases. And the companies that were solving easy problems got some early traction, but then it became really hard for them to scale as the market became super saturated.

### 32:14 — Why Gradial chose workflow over content generation

**[32:14]** That was a lesson that I took into building Gradial is that, you know, at the time there was a lot of focus on content creation. That was the thing people were thinking a lot about. I can automate content and images and video.

**[32:27]** And instead we decided to focus on process and workflow, which didn't seem really appealing at the time for many folks, but for us was like, this is what really is hard about work is all of these things I need to do in order to get something out the door.

**[32:47]** All of the approvals and the checks and the different steps and the dependencies and the teams and the agencies and all this stuff that is required to be able to do something simple.

**[32:57]** That's what we want to go solve for, because if we can go solve for that, we're really solving for reasoning at a more fundamental level and we're solving for process and, you know, eliminating the BS work.

**[33:08]** Why would you want to eliminate content from marketing? Like it's, you know, why not eliminate the things that people don't want to spend time on today and then empower them to go focus on the things that matter.

**[33:20]** So that's really like the big lesson that I took away from working with early stage founders is that we saw this frequently is, you know, when we saw founders chasing a really challenging problem, it was, you know, their success rate was above what was expected.

**[33:35]** And when we saw founders chasing a really relatively easier technical problem, their success rate was lower than what we expected. And so that was a big indicator for me and, you know, where founders should focus when they're building.

**[33:51]** I love that framing. And I think it also makes it worth doing. I think you can get more excited about it too. And because, you know, helping founders build their businesses, building one on your own, you're going to run into challenges left and right.

**[34:06]** And if you can really believe in what you're building, and when you see the impact you're making with one of your customers and knowing that you're really driving change, that's the rocket fuel to just keep it going.

**[34:17]** And I think it's really hard to do that if customers aren't that interested in it, or it doesn't seem like it's really changing their lives. I think it's hard to get that motivation to keep punching through.

### 34:28 — What happens next: the premium moves back to humans

**[34:28]** So I'm curious where you think this goes. It seems like there's been such an explosion of AI capabilities and then this need to really harness it in and build the infrastructure.

**[34:41]** What trends do you think, do your best at whatever crystal ball you're using, what trends do you think will will happen in the next couple years?

**[34:50]** Yeah, I'd say like, there's been a big, you know, focus on like the agent to agent communication and things like that like everyone will have their own agent it'll interact with brand agents and go do tasks for you.

**[35:03]** I increasingly think something like the opposite is going to happen, which is that people are increasingly fatigued with AI and increasingly seeking and placing a premium on human interactions.

**[35:15]** And so I think we're going to start to see something a little bit counterintuitive in that, you know, there's actually going to be a premium for the ability to access or work with a human on things that are important, you know, so if assuming I have the resources,

**[35:32]** things like shopping or travel or financial advice, like things that we would have said are ripe for automation or agent to agent type of things.

**[35:40]** I think we're actually going to see something like the opposite where we're looking for ways that we can spend more time talking to people and less time trying to automate things with AI.

**[35:52]** Now, that creates more opportunity on the B2B side of, hey, well, I want to automate all the painful logistical operational work of a financial advisor or a travel advisor or a nurse or a marketer.

**[36:07]** And so that that's where I think the world is going to trend towards is something similar to what we're seeing in marketing, which is operations and like routine tasks are going to be done with AI and then the actual kind of the face of the brand, the content, the things that are, you know, relationship building,

**[36:28]** things that matter when it comes to taste or helping define the strategy or the future, whether it's for like a trip I want to take or my financial future.

**[36:37]** I think that's actually going to create even more opportunity and work for people.

**[36:44]** And then we'll start to see something like, you know, whereas previously I might have been able to support 10 clients tomorrow, I could support 500 because all of my back office and operational work is done by AI.

**[36:57]** And I think in the past, it's been something like the opposite of like, how do I automate this job or this role or this function or this, you know, the travel agent itself is an AI.

**[37:08]** I don't think, I don't think like everyone wants the travel agent to be an AI I don't I don't think everyone wants that to be a chatbot anymore.

**[37:16]** So now we're going to start to see this countervailing trend that puts more pressure on the technology to go solve the complex, the operational, the workflow, and then actually enables and empowers people to be the front door and relationship holder of that experience.

**[37:31]** So I think that's a, that's a more rewarding and fruitful place to be in general, then, you know, that some of the trends we've seen in the past year where it's like, try to automate everything gear, like I said before that's clearly caused fatigue and, you know, people

**[37:46]** aren't that happy to see that the AI content everywhere they go.

**[37:51]** I know this might sound like an off metaphor for this but when people have talked about this one I agree to the area where I feel like we've seen it is in food.

**[38:03]** So food became increasingly faster, easier, more convenient.

**[38:09]** At the same time ultra processed. And then there was just a massive shift like no I'm not eating another process thing like I need real vegetables, real meat.

**[38:21]** And then the whole push to organic and I will pay double for it like I don't care how affordable it is.

**[38:27]** And I think you're right. I think there's a lot of things where it's like, I don't care if I have to pay more for it, but I want to know that a human is verifying this stuff.

**[38:37]** That's a really good, a really good analogy and that, you know, reducing the varied entry and making things cheaper is not necessarily like yielding the highest quality of output and it really the quality.

**[38:51]** The premium is being put on that quality. When it comes to what what people are really looking for.

### 38:59 — A harness for every function

**[38:59]** You approach this, the enterprise harness really focus on on marketing early in the conversation you mentioned.

**[39:07]** Hey, I think there's going to be opportunities for these verticalized platforms harnesses.

**[39:16]** Where do you see other potential opportunities where something you're working on can similarly be produced for another function or another type of company.

**[39:27]** Yeah, I'd say there's opportunities for every business function. Right.

**[39:31]** Like what we're seeing is that there's, you know, the LMS are fundamentally very horizontal.

**[39:36]** They're enabling like single player productivity to the max. That doesn't mean that they can or should do every function or any function like they they're, you know, if you try to put the LM and the task of being a marketer, then you get the problem with the content being slop, etc.

**[39:53]** So how do we go and enable like the function specific learning to be a part of what the AI is doing the way that a function operates is fundamentally different than another function or someone in a different part of the business.

**[40:07]** And we think that there's a lot of opportunity for like function specific products that allow for a unified experience to connect with and talk to all the tools and agents to accomplish the hero workflows of that function.

**[40:21]** And I think we're starting to see that happen with legal with support with sales and with marketing. I think there's opportunity in every function to do something similar because it's really like a control tower.

**[40:32]** You know, the reality is that it's easy to build agents. It's easy to buy agents, which means that you need a control tower that holds all the context that understands the workflow very well, the approvals and the guardrails very well and understands your organization and the function very well.

**[40:46]** Because then if you have that you can plug in best of breed agents and LLMs and intelligence and people everywhere and you get the best output because you get this organized harness that can orchestrate work.

**[41:01]** Whereas if you just throw agents at the problem or you just throw people at the problem or you just throw tokens like none of those is going to converge to the outcome that the organization needs.

### 41:12 — Wrap

**[41:12]** 100%. And I really like the way you laid it out Anish, I really appreciate being on the podcast because I think your approach and perspective is really, really enlightening for myself and I know a lot of people listening.

**[41:27]** And I think some of the things you mentioned, hey, how can you control the spend, take advantage of whatever model has the best capability for whatever job you're trying to get done, manage the spend, manage the governance of the data who has access to data, who doesn't have access to data.

**[41:45]** And then also just get people focused on the most human part of the work. I think that's a lot of what a lot of us have been trying to build for a long time. How do we take away the boring and just let them do the creative part or the part that requires accountability in the job and trust building and relationship building.

**[42:04]** I'm really impressed with what you've built. I'm really excited to watch you and Grady will grow. You've already done a fantastic job. Congrats on the Series C and all of the amazing companies that you're working with.

**[42:16]** And I hope you come back to give us an update on how things are going in the future.

**[42:21]** Awesome. Thank you so much, Anthony. Really appreciate you having me on and yeah, appreciate the conversation.


---

_LeanScale Knowledge Hub. Free to quote and cite with attribution to The LeanScale Podcast (https://www.leanscale.team)._