> The AI in your product doesn't have to be a chat box
Chat is the obvious way to add AI, and it demos well. The one I keep finding useful is quieter: AI sprinkled through the product, doing a step the user used to have to think about.
Notes on systems and building products at startups and at scale. Product leader with a founder's experience.
Recent Posts
> Why I moved my PM automations from Claude Code into Linear
Moving recurring PM checks into Linear made them something the whole team could see and change. How I set up the loops, and what it suggests about where agents belong.
> What I learned watching agents use my tools
Building CLIs, APIs, and MCPs for agents works a lot like building for people. Watch them use it, see where they get stuck, and fix that.
> Every shortcut pulls a block
A Jenga rule from an old office: fix a bug without a test, pull a block. Why the follow-up after an urgent fix gets lost, and why it's easier to keep now.
> What scheduling looks like when agents are the user
Calendly turned availability into a link. But links were built for humans. When agents book the meeting, the whole model changes.
> Averages hide what your users feel
Your dashboard says average latency is 200ms. Someone is waiting 25 seconds. Averages describe the system. Percentiles describe what your users actually live through.
> SESH - session tracking that survives rabbit holes Blueprint
A session management CLI built on markdown files and git. Here's the architecture, the decisions, and enough to rebuild it yourself.
> Blueprints: open-sourcing the architecture, not the code
Open-source the architecture and the decisions, without the code maintenance. In the era of AI, blueprints are more useful than repos.