- A useful way to think about the agent era is that many products will no longer need to be the place where users do the work; they can instead become the thing that makes the user’s existing agent better. The products that own deep workflows or specialized reasoning may still justify becoming the primary agent, but many others will create more value by exposing their data, context, and actions through agent-friendly surfaces like MCP, APIs, or CLIs. The question is less how do we add an agent? and more where does the user already want the agent to live? tanayj.com/p/build-the-agent-or-power-the-agent
- Maybe Anthropic genuinely believes powerful AI needs tight control, and maybe the safety concerns are real. But the same moves (with Fable) that sound like safety, like keeping usage data, limiting how the API behaves, getting closer to user workflows, and blocking competitors from using the model too freely, also make Anthropic stronger as a business stratechery.com/2026/anthropics-safety-superpower
- Vibe code activity feels like fishing: you get to feel productive sitting on a boat with the line in the water, approving a pull request right after you pull a big one out. It’s also entertaining. stratechery.com/2026/my-vibe-coding-adventure-the-app-and-the-experience-ten-takeaways
- A nice visual guide to learning mahjong. themahjong.guide
What I’ve been reading (W26, 2026)
Agents as a distribution strategy, Anthropic's safety angle, vibe coding as fishing