Essay10 min read
Beck, Karpathy, and Willison Are Saying the Same Thing
Beck, Karpathy, and Willison arrived at the same conclusion independently: AI agents need persistent, structured specifications. The vocabulary differs. The diagnosis doesn't.
Insights
Thinking on software identity, agentic development, and the engineering decisions that persist.
Essay13 min read
Platform Engineering Meets AI Agents: A Governance Framework
Platform teams are the natural governance layer for AI agents. Here's the framework for making it work.
Essay11 min read
Stop Feeding Your AI the Whole Haystack
More context isn't the answer. The right context is. Your agent needs a schema to query, not a haystack to search.
Essay13 min read
The Missing Layer Between Your AI Agent and Your Codebase
Every AI coding agent has context about your code. None have persistent understanding of what your software is supposed to be. That's the gap.
Essay11 min read
Your AI Just Installed a Package That Doesn't Exist. Someone Else Made Sure It Does.
AI models hallucinate package names 20% of the time. Attackers register those names on npm and PyPI. The supply chain attack vector nobody saw coming.
Essay10 min read
Your AI Wrote That Function Three Times. It Thinks They're All New.
AI-assisted coding is linked to 4x more code cloning. Refactoring collapsed from 25% to under 10%. Your codebase has three patterns for the same operation.
Essay11 min read
AI Is Training Your Junior Devs to Be Permanently Junior
Junior dev employment is down 20%. Anthropic's own study shows 17% lower comprehension with AI assistance. The skill pipeline is drying up.
Essay11 min read
Your AI Has Root Access. Your Spec Doesn't.
Replit deleted a production database. Amazon's Kiro agent caused a 13-hour outage. Your AI agents have real permissions with no declared boundaries.
Essay11 min read
Vibe Coding Is Technical Debt on Autopilot
Collins Dictionary's Word of the Year. Y Combinator's favorite approach. And the fastest way to build software you can't maintain.