Hey friends, no full post this week period, but maybe it's worth sharing that I've started building out a public version of the Product AI second brain I use at Cloudflare, and posted it on Github. I imaging this might have limited usefulness since I had to sanitize so many of the internal references that make it truly useful. But maybe—hopefully—there are some things here you can steal and adapt!

Onward, to the links.

Link Posts

• When Using AI Leads to "Brain Fry" — HBR research confirms that cognitive exhaustion from AI oversight is real, and that productivity actually dips after using more than three AI tools simultaneously.

• AI Should Help Us Produce Better Code — Simon Willison and Mitchell Hashimoto both argue that shipping worse code with agents is a choice, and that the fix is engineering your process so agents can't repeat mistakes.

• On Meeting Your Child Again, and Again — Derek Thompson on how parenting is really a permanent relationship with strangers, plural, as your child becomes a new person every few weeks.

• Why It's Still Valuable To Learn To Code — Carson Gross makes the case that if you can't read code you'll fall into the Sorcerer's Apprentice trap, and that senior engineers who stop coding entirely risk the same brain rot they warn juniors about.

• An AI Wake-Up Call — Matt Shumer's argument that AI is different from previous automation waves because it's a general substitute for cognitive work, and the escape routes that existed before are closing as fast as they open.

• Toolshed, Blueprints, and Why Good Agents Need Good DevEx — Stripe's "Minions" series reveals three patterns worth stealing: blueprints that mix deterministic and agentic steps, a centralized MCP server sharing 500 tools across hundreds of agents, and sub-second pre-push lint fixes that benefit humans and LLMs alike.