Field notes on agent-accelerated delivery

How I ship production software with a Claude Code agent fleet — and the senior review layer that catches what the agents miss. From 10+ years in production.

Featured

The Shell-Injection Bugs My Agents Wrote (and the Tests Passed)

Three real shell-injection vulnerabilities coding agents wrote, that automated tests passed, and that a senior human caught before merge — proof AI code needs adversarial review, not just green CI.

aisecuritycode-reviewclaude-code

July 1, 202610 min read

What AI Coding Agents Get Wrong: A Field Taxonomy

Six recurring failure categories I catch reviewing AI-generated code — fail-open security gates, scope creep, silent error swallowing, and more. All pass tests. Here's what to look for.

aicode-reviewsecurityclaude-code

July 8, 20269 min read

2-3x, Not 100x: The Honest Number for AI Coding Speed

How much faster is AI coding really? The measured answer is 2-3x, not 10-100x. Here's the methodology — 30 reviewed PRs in 2.5 days — and what the multiplier leaves out.

aideveloper-productivityclaude-codesoftware-engineering

June 24, 20269 min read

30 PRs in 2.5 Days: Anatomy of Nine Agent Batches

The mechanics behind 30 reviewed PRs in 2.5 days: parallel git worktrees, issue dispatch via CLAUDE.md, and why review — not generation — is the real bottleneck.

aiclaude-codegit-worktreesdeveloper-productivity

June 17, 20269 min read

Claude Code Changed How I Ship Software

How I went from writing every line to orchestrating AI agents — and why 10+ years of engineering experience matters more now, not less.

aiclaude-codedeveloper-productivitysoftware-engineering

March 30, 20265 min read