Aakash Gupta

I stole the AI product stack of the top 1% product managers for you (full tutorial)

Rachel Wolen is CPO at Webflow ($4B company). She runs her whole day out of Claude Code and Cursor. She reveals how to build an agentic Chief of Staff (calendar, email, analytics agents), scale AI adoption from 0% to 30% with Builder Days, and ship AI-native features with MVO before MVP and proper evals. Full Writeup: https://www.news.aakashg.com/p/rachel-wolan-podcast Transcript: https://www.aakashg.com/ai-product-leadership-rachel-wolan-claude-code-cursor/ ---- Timestamps: 0:00 - Intro 1:30 - What Is IC CPO? 3:25 - Building Agentic Chief of Staff 7:03 - Calendar Agent Demo 9:30 - Email Triage Agent 13:01 - Linear Ad Start 13:48 - Analytics Agent (Snowflake + MCP) 17:27 - Linear Ad Start 19:26 - Building Agent from Scratch 29:16 - Setting Up Your Org for AI 34:25 - Shipping AI-Native Features 36:12 - Evals Story 40:58 - Distribution First Mindset 44:00 - Outro ---- 🏆 Thanks to our sponsor: Linear: Plan and build products like the best - https://linear.app/partners/aakash ---- Key Takeaways: 1. IC CPO means self-serving answers - "As a leader, you are able to get your own answers to practically any question." No waiting on data scientists. No back-and-forth with analytics. You have tools to self-serve insights, make analysis, automate workflows. Model behavior for your team to inspire them. 2. Calendar agent analyzes time - Runs weekly with prompt: "Analyze my calendar for last two weeks. Where could I delegate?" Returns delegation opportunities, red flags (double bookings, context switching), what to cut next week. Rachel gives output to EA. Spot on when shown live. 3. Email agent watches behavior - Complete inbox access. Runs triage, archives junk (calendar notifications, marketing), pins important messages, creates draft replies. Twist: watches behavior. If email sits too long, it notices. Caught meeting missing link. Rachel's rule: agent recommends, she approves. No autonomous sending. 4. Analytics agent via MCP - Connected Claude Code to Snowflake via MCP servers (not officially supported repos, just fed them to Claude Code). Ask natural language questions, get SQL executed real-time. "How many sites does Shirts.com have?" Claude writes query, authenticates via SSO, returns answer. Data scientist in pocket. 5. Accept the adoption curve - Your org follows standard curve: early adopters, early majority, late adopters, laggards. Create pathways for everyone to ascend ladder at their pace. Don't force everyone to be you. Rachel to team: "I only want to see prototypes when you have meetings with me." Creates culture investing in prototype quality. 6. Builder Days strategy - Give everyone access: Claude Code licenses, MCP to Snowflake/Tableau, Figma Make, Cursor with design system. Run Builder Days where champions help others through technical hurdles. Everyone demos something outside comfort zone. Results: 0% to 30% of designers using Cursor weekly after first Design Builder Day. 7. Rewrite career ladder - Webflow rewriting career ladder to make AI-native work an expectation, not nice-to-have. Create right incentives. Make sure people supported. Avoid AI for AI's sake. Example: Two designers built similar prototypes. Director caught early: "Go harmonize your prototypes now." Easier now than late in product cycle. 8. MVO before MVP framework - Most teams: Feature → PRD → Design → Ship. Rachel flips it. MVO (Minimal Viable Output) before MVP. Get model's output right FIRST using RAG, prompt engineering, context engineering. Only then build feature. "If you don't have desired outputs, don't spend time productizing the AI feature." 9. Evals are now your job - Brutal story: Webflow's AI app generator 2 weeks from launch. Rachel tested it. Agent kept dying. Realized: changed underlying model, evals didn't have coverage. Evals = test cases for models. Want dream evals (should pass) and edge cases (should fail). Use BrainTrust. Teaching PMs to write evals is part of AI PM toolkit now. 10. Build on your strengths - Framework: See trend → Is it applicable to customers? → What's YOUR core competency? Webflow's strength: bringing visitors to front door via CMS. Built production-grade app generator (not prototype like Lovable). Uses your brand, CMS, hosting, security. "We're bringing a way to prompt an app to production." Don't copy trends, leverage unique strengths. ---- 👨‍💻 Where to find Rachel Wolen: LinkedIn: https://www.linkedin.com/in/rachelwolan/ Twitter/X: https://x.com/rachelwolan Website: https://www.rachelwolan.com/ 👨‍💻 Where to find Aakash: Twitter: https://www.x.com/aakashg0 LinkedIn: https://www.linkedin.com/in/aagupta/ Newsletter: https://www.news.aakashg.com #aipm #claudecode #productmanagement ---- 🧠 About Product Growth: The world's largest podcast focused solely on product + growth, with over 200K+ listeners. 🔔 Subscribe and turn on notifications to get more videos like this.

RWRachel WolenguestAGAakash Guptahost

Nov 30, 2025·45m·Watch on YouTube ↗

📖 CHAPTERS

1. 1
   0:00 – 1:30

   Why AI is becoming the new baseline for product leadership

   Aakash introduces Rachel Wolen (CPO of Webflow) and frames the episode as a masterclass on two tracks: (1) personal productivity as a product leader using AI, and (2) shipping AI-native product features. Rachel sets expectations for live workflow demos and the realities of building an AI-native product close to launch.

2. 2
   1:30 – 3:25

   “IC CPO” explained: getting your own answers as an exec

   Rachel defines the “IC CPO” (individual-contributor Chief Product Officer) as a leader who can self-serve answers to most questions without creating bottlenecks. She outlines the prerequisites: clean/accessible data, the right tools, and modeling experimentation to inspire the team.

3. 3
   3:25 – 7:03

   Building an “agentic Chief of Staff” in Claude Code + Cursor

   Rachel introduces her personal “agentic Chief of Staff,” a growing set of Claude Code agents plus a lightweight app that surfaces their outputs. She explains how she chooses between Cursor, Claude Code, and Codex depending on task complexity and repo context.

4. 4
   7:03 – 9:30

   Calendar agent demo: time audit, prioritization, and delegation cues

   Rachel demos an agent that analyzes her calendar for the last two weeks, identifying delegation opportunities and “red/amber/green” risk flags like double-booking and context switching. The output becomes a practical artifact she can share with her EA to improve scheduling and focus.

5. 5
   9:30 – 13:01

   Email triage agent: archive junk, pin essentials, draft replies (without auto-sending)

   Rachel shows an email agent that classifies inbox noise, recommends archives, pins important threads, and drafts responses for messages that need action. She emphasizes keeping the human in control—reviewing drafts and approving actions—while letting the agent accelerate throughput.

6. 6
   13:01 – 13:48

   How the integrations work: Google tokens, .env hygiene, and safe local execution

   Rachel explains the practical setup for connecting Gmail/Calendar: generating Google Cloud tokens, storing secrets in a local .env, and ensuring the file is gitignored. She also notes the importance of thinking about access control, what data gets exposed to models, and running within enterprise accounts where possible.

7. 7
   13:48 – 17:27

   Analytics agent with Snowflake + MCP: “a data scientist in my pocket”

   Rachel demonstrates querying Snowflake via natural language inside Claude using MCP servers, enabling quick customer/workspace insights without pulling a data scientist into ad-hoc requests. She explains that documentation (e.g., dbt models) materially improves the quality of NL-to-SQL outputs.

8. 8
   17:27 – 19:26

   Organizing and invoking agents: markdown files, separate windows, and an app UI

   Rachel shows how agents are organized as markdown configuration files in an agents folder and invoked through context in Cursor/Claude Code. She also shares why she built a small app: reading raw markdown is awkward, so the app displays agent outputs like podcast prep and guest research in a more usable format.

9. 9
   19:26 – 29:16

   Build an agent from scratch: LinkedIn post generator + meme image pipeline

   Rachel and Aakash create a new agent end-to-end: defining its purpose, granting repo access, and having Claude generate the agent spec. They add reference materials (best posts + writing guidelines), test the agent, and discuss iterative tuning as the real path to reliability.

10. 10
    29:16 – 34:25

    Setting up your org for AI adoption: champions, builder days, and new incentives

    Rachel lays out an adoption strategy grounded in classic diffusion curves: early adopters through laggards. She describes trainings (Cursor, Figma Make), “builder days” with demos, shifting expectations toward prototypes, and even updating career ladders to encode AI fluency as part of performance.

11. 11
    34:25 – 36:12

    Shipping AI-native features: the evals failure that broke a near-launch product

    Rachel shares a real incident: a model swap caused their app-gen product to break because eval coverage wasn’t sufficient to catch the regression. She frames evals as the new “test suite” for AI behavior—hard to design, especially when you need both passing and intentionally failing cases to validate model changes.

12. 12
    36:12 – 40:58

    Choosing AI features that match your strengths: “prompt an app to production”

    Rachel explains how Webflow positions its AI app generation around core strengths: brand-consistent outputs, CMS integration, production-grade hosting/security, and workflows that serve designers, developers, and marketers. The strategic goal is differentiation from prototype-first tools by focusing on production readiness and native integration.

13. 13
    40:58 – 44:00

    Distribution-first mindset: from SEO to answer-engine optimization (AEO)

    Rachel outlines how distribution evolves with tech waves: SEO for the web, app-store optimization for mobile, virality for social, and now visibility in “answer engines” like ChatGPT. She argues product teams must build for discoverability in these new surfaces, including keeping knowledge current via site content (FAQs) and preparing for agentic browsers that interact with websites/apps directly.

14. 14
    44:00 – 45:42

    Wrap-up: the new roadmap for top-tier AI product leaders

    Aakash recaps the episode’s playbook: become an IC CPO with agentic workflows, enable your org through access/training/incentives, and ship AI-native products with eval rigor, strategic differentiation, and distribution-aware thinking. The episode closes with pointers to additional resources and links.