Writing
Thoughts on development, tooling, and the messy reality of building things.
-
GitHub Breach Shows Microsoft's Supply Chain Problem
GitHub breach shows how Microsoft's developer tools turned trusted extensions, npm workflows, and auto-updates into attacker delivery paths. Read why.
-
Google Developer Trust and the Cost of Gemini 3.5 Flash
Google developer trust is cracking as Gemini 3.5 Flash pricing, the Antigravity CLI shift, and the Railway outage expose the cost of relying on Google.
-
HTML vs Markdown: The Agent Output Debate Is Wrong
HTML vs Markdown is the wrong fight. Learn when AI agent output should be durable text, rendered HTML, or throwaway interfaces that help you decide well.
-
OpenAI Azure Breakup: How Anthropic Forced the Split
Anthropic's Bedrock distribution forced OpenAI to break Azure exclusivity. Inside the deal collapse, the leaked Dresser memo, and the $100B AWS pivot.
-
AI Token Pricing: The Free Ride Is Over for Good
AI tokens are getting cheaper per unit, but total enterprise spending keeps climbing. GitHub Copilot's billing shift and Uber's budget blowout show why.
-
GitHub Merge Queue Bug: What Happened and Why It Matters
GitHub's merge queue silently corrupted 2,092 pull requests across 658 repos on April 23, 2026. A faulty feature flag, no automated detection, and a trust contract broken.
-
Goodhart's Law Ate the AI Industry
Benchmark scores, chart lines, token counts, GitHub stars — every metric the AI industry uses to prove progress is being gamed, bought, or distorted. Not a single one can be trusted.
-
Why You Can't Stop Opening That App
Duolingo gets credit for gamification, but the apps with 90%+ retention aren't decorating with badges. They're running craving machines, infinite games, and invisible scoreboards.
-
Trust Us, It's Dangerous
Anthropic says Claude Mythos is too dangerous to release. Smaller models reproduce its flagship results. The $100M in 'usage credits' goes to Anthropic's own platform. And the IPO is in Q4.
-
Anthropic Got Caught
Anthropic promoted an iMessage plugin that breaks Apple's ToS while simultaneously threatening legal action against Open Code for doing the same thing to their platform.
-
Bash Was a Happy Accident
Bash became the default execution layer for AI agents because it was there, not because it was right. TypeScript isolates are what these agents actually need.
-
I Made Antigravity Switch Its Own Brain
Antigravity is an Electron app. Electron is Chromium. Chromium has a debug backdoor. I used it to make the IDE swap its own AI model mid-workflow.