Field Guide

What AI coding agents actually do

whybother ships ten features. Each one is a real behaviour of AI coding agents, played straight. This is the other half: what is actually happening, how it shows up in a diff, and the guardrail that catches it before it ships.

None of this is about a particular model. Models change every quarter; the incentives that produce these behaviours do not. An agent told to make the tests pass will find the cheapest way to make the tests pass.

• Test OptimizationWhy AI coding agents delete failing testsGreen on merge. Suite: none required.
• Self-Healing ErrorsWhy AI agents swallow errors with try/catchEvery error handled. Every handler returns null.
• Instant DependenciesWhy AI agents add dependencies you never needed1,847 packages, some of which exist.
• Zero-Config SecretsWhy AI agents hard-code API keysThe key is in App.tsx. Rotation is included.
• Production-Ready by DefaultWhy AI agents call it production-ready when it is notDeployed. White screen. Zero known bugs.
• Proactive FeaturesWhy AI agents add features nobody asked forOne fix. Five improvements. A globe.
• Always AgreeableWhy AI agents agree with you when you are wrongYou’re absolutely right. Also, a different plan.
• Infinite ContextWhy AI agents forget your requirementsRequirements compacted. Mood preserved.
• Clean Slate RecoveryWhy AI agents run git reset --hard and delete .gitHistory: optional. State: clean.
• One-Click ReviewWhy AI-generated code gets approved without reviewReviewed 4,812 lines in 0.3s. Merged.

Short on time? The checklist fits on one page. Want the long version? The reading list is older than every agent and still right.

Want your README to confess, or to brag? There are badges, one pair per entry.