008 entries
001 2026.09.02
September 2, 2026
AI recommendation is a distribution channel that never reports back. I evaluate my DevRel program the same way I evaluate an agentic workflow, checking presence first, then accuracy, and auditing both on a cadence.
002 2026.07.29
July 29, 2026
A re-audit of Stripe, Clerk, and Neon shows the best developer-tool docs now reach five of six AEO layers. That clarifies who owns AI-readable infrastructure.
003 2026.07.09
July 9, 2026
Durable AI workflows are found through iterative rebuilding, not upfront design. The framework emergence loop turns each failure into the next version's lesson.
004 2026.06.18
June 18, 2026
A broken agentic workflow fails loudly. The wrong one runs cleanly and quietly drifts. How to define correctness and evaluate against it before you ship.
005 2026.05.18
May 18, 2026
AI-assisted builds stall when the specification leaves decisions undefined. The specification boundary is the missing territory you must plan before the model can finish.
006 2026.05.11
May 11, 2026
The reason AI-assisted builds stall isn't the model: design decisions were never made. Here's why writing the spec first changes everything.
007 2026.05.04
May 4, 2026
An LLM context file is a maintained deliverable, not a config toggle. It tells an agent what the project is, what decisions stand, and how to navigate it.
008 2026.04.27
April 27, 2026
The AX shift means designing docs, context files, and specs for agents. UX serves users; DX serves engineers; AX serves the agent as primary reader.
stay in touch
I write and connect with operators building developer-first products and payment systems. Follow along on LinkedIn. That's where conversations happen.
the newsletter
Occasional writing on developer platforms, payments, and AI-assisted engineering. No spam, and you can unsubscribe anytime.