Why Context Windows Won’t Solve AI Agent Memory
Why long-running agents need selective, versioned memory — with retrieval, provenance, and a way to forget stale information — even as context windows grow.
Articles
Writing about infrastructure, developer platforms, agent memory, and the decisions behind them.
An AI Agent Without Context Is a Fighter Jet Without Radar
Why capable agents still fail inside specialized domains — and what changes when context becomes structured, inspectable infrastructure.
Your Build Isn’t Slow. Your Runner Queue Is.
A look at the hidden wait time in CI, why faster compilation does not fix it, and the trade-offs behind runner infrastructure.
Why Most Internal Developer Platforms Fail
Great engineering is not enough. Internal platforms work when teams treat adoption, golden paths, and developer trust as product problems.
World models are the next AI race. Infrastructure is missing from it.
What infrastructure could learn from world models: moving beyond dependency graphs to simulate failures, predict cascades, and explore what happens next.
Bazel 9 Migration: How to Get Faster Builds Before the Bzlmod Refactor
A pragmatic argument for separating build-performance improvements from the migration work required by a major build-system release.
Why AOSP Builds Take Forever
Where Android build time actually goes, why adding hardware has limits, and how shared caching changes the economics of compilation.