Writing  /  Topic

Tooling reviews

Honest operational reads on the coding agents I run daily, each on a distinct axis: head-to-head, what breaks at scale, and what twelve months actually cost.

Tooling Review MAR 2026
Claude Code in production: what breaks after 5,000 jobs Five thousand Claude Code jobs on a home server. The model was never the hard part. Rate limits, context bloat, and session state were, and the fixes were all infrastructure. Read · 1 min
Tooling Review MAR 2026
What 12 months of daily AI coding in production actually cost Twelve months running Claude Code, GPT-5.4, Grok, Windsurf, and local Llama daily. What I actually paid, where each tool earned its place, and what the bill teaches you that benchmarks don't. Read · 10 min

Other topics

The harness thesisFailure postmortemsArchitecture deep-divesContext and memory engineeringStart here