I've spent the last week building you a Team OS in Claude Code
This is the unlock most product teams need to get the most out of Claude Code
OpenAI published a line in their harness engineering post in February that I haven’t stopped thinking about:
“That Slack discussion that aligned the team on an architectural pattern? If it isn’t discoverable to the agent, it’s illegible in the same way it would be unknown to a new hire joining three months later.”
They were talking about AI coding agents. But read it again. It applies to your entire product team.
That pricing decision your team made in February? If it’s buried in reply #31 of a Slack thread, it doesn’t exist. Not for the AI. And not for the PM who started last month. That customer call where the VP of Engineering said the integration was a dealbreaker? If the summary lives in one PM’s personal notes, it might as well not have happened. The churn query your analyst wrote last quarter? If the engineer paged at 2 AM can’t find it, she’s stuck until morning.
The knowledge exists. It’s just trapped: in someone’s head, in a personal setup, in a tool nobody else can search.
The Solution: A Team OS
Hannah Stulberg, a PM at DoorDash, showed the solution on my podcast a few weeks back.
She built a shared repo where every team function checks in their context and anyone on the team can query it in natural language. She demoed a query that found the answer using 3% of the context window.
The part that stuck with me was what happened the following week, when a new engineer needed context about a customer decision from three months ago.
Instead of pinging Hannah in Slack and waiting, the engineer opened the repo, asked the question, and got the full reasoning in 15 seconds.
Hannah wasn’t involved. She wasn’t even online.
She had made herself unnecessary for context questions. The team treated her as more valuable for it. Every PM book tells you to make yourself indispensable. Hannah's bet was the opposite. She freed herself from being the bottleneck so she could do the work that actually requires judgment.
I’ve spent months building personal AI setups. My PM OS has 41 skills and a context library and sub-agents. I’m proud of it. And I’ve slowly realized it solves the first problem. It makes me faster. The harder problem is making the team faster. When I’m asleep at 2 AM and an engineer needs the churn query, my personal OS can’t help them.
I’ve spent the past week studying four implementations of what I’m calling a Team OS. Hannah at DoorDash. Dave Killeen at Pendo. Gabor Meyer at Google. Carl Vellotti building solo. Four people, four companies, four different levels of complexity. They all converged on the same architecture.
That convergence is what convinced me to write this.
Today’s Post
I’ve partnered up with Hannah Stulberg to make sure this followup to her podcast is the highest quality possible guide to building a Team OS in Claude Code out there.
She is the author of In the Weeds (including her viral Claude Code for Everything series) and is running a Maven workshop on the topic this weekend.
Why We Built This
New hires take 6 to 7 months to feel settled.
Only 12% of employees say their company does onboarding well.
47% of companies call institutional knowledge loss their top offboarding challenge.
Most of that time, most of that loss: context that already exists somewhere, trapped in someone’s head or buried in a tool nobody else can search.
A context question is rarely just the lookup time. It's the lookup, the Slack ping, the wait, the context-switch back. At 10 of those a day at 10 minutes of total productive time lost each, you're at over 8 hours a week. Most teams I've talked to are higher.
And the pattern isn’t Claude Code specific. Codex, Cursor, GitHub Copilot. This architecture can be moved wherever you are, because it’s just Markdown files. The examples here just use Claude Code because that’s what is most popular with AI-native PMs right now.
The Deep Dive
🔒 For paid subscribers, we’ve put together everything you need:
How to Upgrade your Personal OS to a Team OS with 1 command
The Architecture That Works (The 3 Layers)
Three Ways Teams Actually Use It
How to Build One in 4 Weeks
The Real Problems You’ll Encounter
What the First Month Actually Looks Like
Plus 6 downloadable resources, including: a skill to convert your personal OS to a team OS, a starter repo, and an adoption playbook.
Keep reading with a 7-day free trial
Subscribe to Product Growth to keep reading this post and get 7 days of free access to the full post archives.
 | A guest post by
|