Managing Context Windows and Compaction

Today’s episode

There is a term Andrej Karpathy coined last year: vibe coding.

We have the same for product management: Vibe PMing.

You describe the problem. The agent pulls the data. Analyzes the chart. Synthesizes the feedback. Drafts the spec. Files the ticket.

That is not theory. That is what I walked through in today’s episode with a principal PM at Amplitude who builds MCP and agent products for a living. He showed it live, on screen, in real time.

If you tune in, you’ll learn the full end-to-end workflow:

----

If you want access to my AI tool stack - Dovetail, Arize, Linear, Descript, Reforge Build, DeepSky, Relay.app, Magic Patterns, Speechify, and Mobbin - grab Aakash’s bundle.

----

Check out the conversation on Apple, Spotify and YouTube.

Brought to you by:

* Amplitude: The market-leader in product analytics

* Pendo: The #1 software experience management platform

* Testkube: Leading test orchestration platform

* Product Faculty: Get $550 off the AI PM Certification with code AAKASH550C7

* Bolt: Ship AI-powered products 10x faster

----

Key Takeaways:1. Claude Code + MCP is the most powerful AIPM workflow today - Connect your analytics tool via MCP, load your product context into a repo, and let the agent do analysis that used to take hours in minutes.2. Deep chart analysis now takes 90 seconds instead of 3 hours - Drop a chart URL into Claude Code, trigger the analyse chart skill, and the agent navigates your data taxonomy, finds anomalies, and hypothesises why metrics changed.3. Automate your entire weekly business review - Point Claude Code at your dashboards Monday morning. Get 3-5 top insights and the one urgent issue to tackle — no manual dashboard scanning ever again.4. Customer feedback synthesis across all channels in one pass - Zendesk, Gong, Salesforce, Slack, app stores all unified. Claude Code navigates the MCP, clusters themes, and surfaces what customers love and hate that week.5. PRDs write themselves from insights - Take the analysis output, point it at your PRD template in Cursor or Claude Code, and get a first draft spec in under 2 minutes. Iterate with command L or command K.6. Skills are the most important Claude Code feature - A skill is just a named prompt with heuristics and tool instructions. It loads only when relevant, preventing context bloat and giving the agent a repeatable workflow.7. The biggest MCP mistake is connecting too many servers - Every tool description burns context. Load only what's relevant to the workflow. Remove or hide tools that aren't being used for a given task.8. MCP is not for complex orchestration — it's for data access - Set the right expectation. MCP connects AI to external systems easily. It's the first step, not the whole pipeline.9. Granola has no MCP, so build a script instead - Frank used Claude Code to write a local script that dumps Granola meeting notes into his product repo. Now he can pull all meeting context with a single at-command.10. The future of PMing is vibe PMing - Chart analysis, dashboard reporting, feedback synthesis, spec writing, and prototyping — all agent-driven. PMs who adopt this workflow now will have a massive advantage in 2-3 years.

----

Related content

Newsletters:

* How to use Claude Code like a pro

* Steal 6 of my Claude skills

* Context engineering

* The AI stack for PMs

* Practical AI agents for PMs

Podcasts:

* How to build an AI-native PM operating system with Mike Bal

* AI evals explained simply with Ankit Shukla

* Advanced guide to AI prototyping with Sachin Rekhi

PS. Please subscribe on YouTube and follow on Apple & Spotify. It helps!

This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.news.aakashg.com/subscribe