Aakash GuptaI stole the AI product stack of the top 1% product managers for you (full tutorial)
CHAPTERS
- 0:00 – 2:02
Rachel Wolen’s AI “superpowers”: running a CPO day from Claude Code + Cursor
Aakash introduces Rachel Wolen (CPO at Webflow) and frames the episode as a tactical masterclass for product leaders—not just IC PMs—on using AI tools to work faster and smarter. Rachel tees up her two tracks: personal productivity via agents and shipping AI-native products.
- •Rachel runs her day in Claude Code and Cursor and encourages team-wide adoption
- •Episode focus: AI productivity for leaders + building AI-native features
- •Preview of agents: calendar, email triage, analytics, and more
- •Promise: a practical tutorial, not just high-level theory
- 2:02 – 3:24
What “IC CPO” really means: self-serve answers, tools, and modeling experimentation
Rachel defines IC CPO as the ability for a leader to get their own answers to almost any question. She explains the prerequisites: well-shaped data, the right tools across the team, and leadership that models experimentation to inspire others.
- •IC CPO goal: independently answer questions without bottlenecking on others
- •Work backwards: data quality + self-serve access for the whole team
- •Tooling matters, and teams need a “stair-step” path to level up
- •Leaders should experiment publicly to normalize iteration and failure
- 3:24 – 4:31
Building an agentic Chief of Staff: tool stack choices and the agents folder approach
Rachel outlines her “agentic Chief of Staff,” combining Claude Code agents with a lightweight app she built for daily use. She explains when she uses Cursor, Cursor’s agent, or Codex, and how she continuously adds and refines agents stored as simple markdown files.
- •Chief of Staff = multiple Claude Code agents + a custom app UI
- •Uses Cursor/terminal workflows; sometimes Codex for complex monorepo context
- •Agents are simple, composable, and constantly evolving
- •Agent definitions live as markdown files that Claude can invoke
- 4:31 – 7:09
Calendar Agent demo: auditing time, delegation opportunities, and context-switching flags
Rachel demonstrates a calendar-priority/time-audit agent that analyzes how she spent her time and what to delegate. The output highlights meetings she should skip, double-booking issues, and recommendations she shares with her EA to improve weekly planning.
- •Weekly (and sometimes daily) calendar analysis to improve focus
- •Delegation suggestions based on meeting type and relevance
- •Flags for double/triple booking and excessive context switching
- •Operationalizes results by sending insights to her EA for adjustments
- 7:09 – 8:25
Email triage agent: archive junk, pin priorities, and draft replies (with human approval)
Rachel shows an email agent that cleans up inbox noise, recommends what to archive, and drafts responses to unblock communication—without sending automatically. She emphasizes that agent behavior improves through repeated iteration and correction.
- •First pass: remove newsletters/notifications/system messages
- •Keeps and pins important threads; drafts replies for review
- •Catches issues like meetings missing links
- •Design principle: the agent recommends actions; human approves
- 8:25 – 9:30
How to connect Gmail/Calendar securely: tokens, .env hygiene, and repo safety
Rachel explains the practical setup: generating Google Cloud tokens, storing them in a local .env, and ensuring secrets don’t sync via Git. She describes how Claude Code can scaffold the environment variable setup while the user completes token generation in the console.
- •Create OAuth/token in Google Cloud, store in local .env
- •Keep .env ignored by Git/Cursor to prevent secret leakage
- •Claude Code can scaffold needed env vars and config
- •Regenerate tokens when needed; treat credentials as disposable
- 9:30 – 15:35
Analytics agent with Snowflake + MCP: natural language queries like a “data scientist in your pocket”
Rachel demonstrates querying Snowflake via Claude using MCP servers, enabling quick customer/workspace insights without pulling a data scientist into every request. She highlights that good results require documented models (e.g., dbt) and that access is governed by existing SSO permissions.
- •Ask natural language questions; agent writes and runs Snowflake SQL
- •Use case: quickly understand what a customer is using before a meeting
- •Model documentation (dbt) improves accuracy of NL-to-SQL
- •MCP servers for Snowflake/Tableau; runs locally with SSO-based access
- •Privacy/security lens: don’t grant models more than you already can access
- 15:35 – 17:22
Organizing and invoking agents: multi-window workflows and a personal “agent outputs” app
Rachel explains how she invokes agents by keeping markdown agent files in an agents folder within the repo; Claude picks them up from context. She also built a UI app that renders outputs (calendar, prep docs) more readably than raw markdown.
- •Invocation: Claude recognizes the right agent from the agents folder context
- •Separate windows help keep tasks isolated and reduce confusion
- •Custom app displays agent outputs and daily context (calendar, docs)
- •Examples: podcast prep researcher, video transcriber, dinner guest research
- 17:22 – 19:26
Sponsor break: Linear as the fast product dev hub with AI agents
Aakash reads a sponsor segment explaining Linear’s positioning as a faster alternative to Jira and a product development hub. He highlights Linear’s AI agents and automation for PM work like synthesis, PRDs, scoping, and status updates.
- •Linear positioned for speed; used by leading AI-native teams
- •AI agents can help execute tickets with human review before production
- •Product intelligence surfaces insights from feedback/support tickets
- •Pitch and call-to-action link shared by host
- 19:26 – 22:57
Building an agent from scratch: LinkedIn post generator + image creation workflow
Rachel live-builds a new agent in Cursor: a LinkedIn post generator that also produces a meme image via OpenAI’s image model. They walk through creating an agent config, granting project access, generating the markdown spec, and reviewing it with markdown preview.
- •Create a new agent via ‘manage agent configurations’
- •Define tasks: analyze reference materials, draft post, generate meme image
- •Use Claude (Sonnet) to generate the agent’s markdown instructions
- •Cursor tip: open markdown preview for easier reading
- •Small stylistic constraint example: avoid em dashes to reduce “AI tells”
- 22:57 – 29:16
Feeding reference materials & iterating: drag-and-drop files, test runs, and tuning over time
They add reference posts and guidelines by dragging files into the terminal, then update the agent to use them. Rachel tests the agent on a Halloween prompt, discusses the importance of iteration, and explains how new model releases can improve tool-calling and long-running workflows.
- •Drag files into terminal to add reference materials quickly
- •Verify the agent is actually referencing uploaded materials
- •Test output: a usable LinkedIn post + attempt at image generation
- •Key practice: assume version 1 is step 1 of 3–4; tune with real usage
- •Model upgrades can materially improve agent reliability and tool use
- 29:16 – 34:35
Setting up an AI-native product org: adoption curves, training, builder days, and incentives
Rachel outlines an organizational playbook: expect different adopter types, make tools accessible, and create support structures that help people climb the learning curve. Webflow runs trainings and builder days, uses internal champions, and is even rewriting the career ladder to reinforce AI fluency—while staying outcome-focused (not AI for AI’s sake).
- •Plan for early adopters through laggards; tailor enablement accordingly
- •Behavior shift: ‘show prototypes in meetings’ changes team operating rhythm
- •Enablement: Cursor/Figma Make trainings, builder days, champion-led support
- •Measure adoption impact (e.g., design team weekly usage jump)
- •Update incentives: incorporate AI capability into career ladder expectations
- •Outcome focus: prototypes help harmonize workflows earlier in the lifecycle
- 34:35 – 38:04
Shipping AI-native features: the evals lesson from a model swap that broke the product
Rachel shares a real incident: Webflow’s app gen product broke after changing the underlying model because eval coverage was insufficient. She explains evals as test cases for model behavior, the difficulty of creating “dream evals,” and how PMs must learn this emerging discipline, including synthetic evals and tooling like BrainTrust.
- •Model swaps can regress behavior unless evals catch failures
- •Evals = test cases; need both expected-pass and expected-fail scenarios
- •Coverage gaps are common and hard to foresee without discipline
- •PM enablement challenge: teaching eval writing and maintaining coverage
- •Tooling/partners: BrainTrust; consider synthetic eval generation
- 38:04 – 41:04
Choosing AI features that match your strengths: Webflow’s differentiated ‘prompt to production’ approach
Rachel explains how Webflow selects AI roadmap bets by mapping market trends to customer needs and leaning into platform strengths: CMS, design systems, hosting, security, and brand-consistent outputs. The goal is not a prototype generator, but a production-grade app workflow that fits a wide range of users.
- •Start from customer-relevant market trends, not hype
- •Differentiate via native platform advantages (CMS, components, design system)
- •Production-grade principles: brand consistency, workflow integration, hosting/security
- •Serve a broad persona set: designers, developers, marketers
- •Positioning: ‘prompt an app to production,’ not ‘generate a prototype’
- 41:04 – 43:44
Distribution-first mindset: from SEO to AEO (answer engine optimization) and agentic discovery
Rachel describes distribution as evolving waves: SEO (web), ASO (apps), virality (social), and now answer engines. She argues teams must adapt product and website processes (FAQs, updates, structured info) to be discoverable and correctly represented in systems like ChatGPT and future agentic browsers.
- •Distribution waves: SEO → ASO → social virality → answer engines
- •AEO: get your brand/products correctly surfaced in answer engines
- •Operational implication: keep website/FAQ content current and machine-legible
- •Anticipate agentic browsers/apps interacting with websites and stores
- •Growth opportunity: discovery is shifting; build for the new channels early
- 43:44 – 45:42
Wrap-up: the roadmap to top 1% AI product leadership + where to get resources
Aakash summarizes the full arc: IC CPO productivity workflows, building and organizing agents, and org-level enablement; then the core lessons for AI-native shipping (evals, strengths-based differentiation, distribution). They close with plugs for the newsletter and where to find the linked resources and documents.
- •Recap: agents for calendar/email/analytics + building agents from scratch
- •Org strategy recap: access, support, incentives
- •Shipping recap: evals, play to strengths, distribution-first thinking
- •Pointers to newsletter post for links/docs and podcast platforms
- •Final CTAs: subscribe/follow/review; Webflow app builder mention
