Aakash GuptaI stole the AI product stack of the top 1% product managers for you (full tutorial)
Aakash Gupta and Rachel Wolen on webflow CPO’s AI agent stack for elite product leadership today.
In this episode of Aakash Gupta, featuring Rachel Wolen and Aakash Gupta, I stole the AI product stack of the top 1% product managers for you (full tutorial) explores webflow CPO’s AI agent stack for elite product leadership today The “IC CPO” mindset is framed as a leader’s ability to self-serve answers by investing in high-quality, accessible data and modern AI tooling across the team.
At a glance
WHAT IT’S REALLY ABOUT
Webflow CPO’s AI agent stack for elite product leadership today
- The “IC CPO” mindset is framed as a leader’s ability to self-serve answers by investing in high-quality, accessible data and modern AI tooling across the team.
- Wolen demos an “agentic Chief of Staff” stack that analyzes calendar time, flags conflicts, recommends delegation, triages email with draft replies, and supports meeting/dinner research.
- She shows an analytics agent that uses MCP servers to query Snowflake/Tableau via natural language, enabled by documented dbt models to improve query accuracy and reliability.
- A live walkthrough illustrates creating a new agent (LinkedIn post generator) by generating a markdown agent spec, attaching reference files, and iterating through trial-and-adjust cycles.
- On shipping AI-native products, she highlights eval coverage as a critical failure point when swapping models, argues for building AI features aligned to company strengths, and urges a distribution-first approach focused on “answer engines” (AEO) as the next discovery wave.
IDEAS WORTH REMEMBERING
7 ideasDefine “IC CPO” as self-serve answers, then backsolve the infrastructure.
Wolen argues leaders should be able to answer most questions directly, which forces investment in clean data, self-serve access, and tools that let non-specialists retrieve insights without constant analyst dependency.
Start with high-leverage personal agents: calendar and inbox triage.
Her calendar agent reviews time allocation, flags overbooking/context switching, and suggests delegation; her email agent bulk-archives noise, pins important messages, and drafts replies—while keeping humans in control of sending.
Treat agents as iterative products, not one-shot automations.
She emphasizes running agents, observing errors, and tightening instructions; most agents take “three or four” iterations to dial in, especially for critical workflows like email drafting or analytics.
Analytics agents work best when your data models are documented.
Natural-language querying against Snowflake improved meaningfully once dbt models were well documented, enabling more accurate SQL generation and reducing back-and-forth to interpret schemas.
Use MCP servers to connect enterprise tools while preserving access controls.
She sets up MCP for Snowflake/Tableau using SSO credentials so the agent only queries data she already has permission to access, and she stresses privacy considerations in what context gets exposed to models.
Operationalize AI adoption with access, champions, and incentives.
Webflow expanded usage via trainings (Cursor/Figma Make), “builder days” where everyone demos something, champion support to clear technical hurdles, and even rewriting the career ladder to encode AI expectations.
Evals are the new regression tests—lack of coverage will bite on model swaps.
A near-launch app-gen workflow broke after changing the underlying model because evals didn’t fail appropriately; she frames eval design (including synthetic evals) as a core modern PM skill and mentions BrainTrust as their platform.
WORDS WORTH SAVING
5 quotesTo me, IC CPO means that as a leader, you are able to get your own answers to practically any question.
— Rachel Wolen
It’s almost like having a data scientist in my pocket.
— Rachel Wolen
A lot of building an agent is trying it out and then going and adjusting what you want the agent to do.
— Rachel Wolen
We had changed the underlying model, and our evals didn’t have enough coverage to fail when we changed the model.
— Rachel Wolen
The next wave is really going to be around how do you get listed in answer engines?
— Rachel Wolen
QUESTIONS ANSWERED IN THIS EPISODE
5 questionsWhat specific signals or rules does your calendar agent use to decide a meeting is “delegatable” versus truly CPO-critical?
The “IC CPO” mindset is framed as a leader’s ability to self-serve answers by investing in high-quality, accessible data and modern AI tooling across the team.
How do you prevent email and calendar agents from creating security/compliance risk (e.g., handling sensitive threads) while still being useful?
Wolen demos an “agentic Chief of Staff” stack that analyzes calendar time, flags conflicts, recommends delegation, triages email with draft replies, and supports meeting/dinner research.
What did you learn from the model swap that broke the app-gen product—what evals did you add afterward to prevent a repeat?
She shows an analytics agent that uses MCP servers to query Snowflake/Tableau via natural language, enabled by documented dbt models to improve query accuracy and reliability.
How do you decide when to use Claude Code vs Cursor’s agent vs Codex for different tasks (speed, repo complexity, reliability)?
A live walkthrough illustrates creating a new agent (LinkedIn post generator) by generating a markdown agent spec, attaching reference files, and iterating through trial-and-adjust cycles.
Can you share a concrete template for an agent markdown spec (role, tools, constraints, failure handling) that consistently works for you?
On shipping AI-native products, she highlights eval coverage as a critical failure point when swapping models, argues for building AI features aligned to company strengths, and urges a distribution-first approach focused on “answer engines” (AEO) as the next discovery wave.
EVERY SPOKEN WORD
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome