Guides

Shipping AI-written code that survives review.

Field guides for engineers using Claude Code, Cursor, Windsurf and other AI assistants — on the failure modes that matter and the gates that catch them.

Jul 2026 · 7 min

How to catch AI-hallucinated code before it ships

Invented SDK calls, faked refund paths, mocks in the wire path — the four shapes hallucinated code takes, and the two-pass gate architecture that catches them.

Read

Jul 2026 · 6 min

How to add a code-review gate to Claude Code with an MCP server

The five-minute setup: wire a veto-empowered validation server into Claude Code, Cursor or Windsurf, and make gating the default with a standing project rule.

Read

Jul 2026 · 7 min

Why AI coding assistants miss deep performance bugs

N+1 storms, scalability cliffs, blocking hot paths: why self-review can't see them, and what independent review with production arithmetic looks like.

Read

Jul 2026 · 6 min

Use smaller, cheaper AI coding models — safely

Enforce quality at review instead of assuming it at generation, and the choice of model becomes a cost decision, not a risk decision. A three-week migration path.

Read