Avtar Varma, who just started as VP RevOps at Chainguard after building RevOps functions from Series A startups to Stripe, lays out three stages of enterprise AI adoption: token maxing, centralizing within functions, and a company-wide AI ops team. He explains why usage leaderboards are a vanity metric and what to reward instead. Together with Janis and Philipp, he debates where an AI ops function should sit, who owns the data infrastructure, and why go-to-market is the best place to prove ROI. He closes with an unpopular take on go-to-market engineers.

We cover:

- Stage one: token maxing, and why "give everyone every AI tool" makes ROI hard to prove

- Why AI usage leaderboards are a vanity metric

- Hackathons vs. leaderboards: rewarding the best ideas, not the biggest spend

- Meeting prep as a simple, high-impact use case, and why AI has to live in the calendar

- Stage two: centralizing skills, templates and agents inside each function

- Using usage data to coach below-average adopters instead of rewarding top spenders

- Stage three: a central AI ops team with a liaison for each function

- Where AI ops should sit: COO, CIO or RevOps, and who owns data infrastructure

- Saving cost with open-source models where best-in-class models aren't needed

- Why the data warehouse is becoming the layer agents run on

- Whether RevOps can grow into BizOps or AI ops

- Why go-to-market is the best place to start proving AI ROI

- Unpopular opinion: go-to-market engineers should be actual engineers

Avtar Varma: https://www.linkedin.com/in/avtar-varma

Chainguard: https://www.chainguard.dev/

Janis Zech: https://www.linkedin.com/in/janiszech/

Philipp Stelzer: https://www.linkedin.com/in/philippstelzer/

WeFlow: https://www.weflow.ai/

Community: https://www.weflow.ai/community

RevOps Letter: https://www.weflow.ai/revops-letter

Chapters:

(00:01) Avtar Varma joins from Chainguard

(01:58) Three stages of AI adoption

(05:54) Is token maxing necessary?

(07:33) Leaderboards as a vanity metric

(10:33) Hackathons and meeting prep

(14:46) Incentives that actually work

(16:48) Coaching with usage data

(18:10) Stage three: central AI ops

(21:07) Who owns data and infrastructure

(22:17) Cost, open-source models, security

(24:46) Can RevOps become AI ops?

(27:46) Why GTM is the place to start

(28:31) Unpopular opinion: GTM engineers