A practical, opinionated manual for using AI coding agents with production-grade engineering discipline.
The biggest risk of AI coding isn't bad code - it's loss of agency. A gradual surrender of control over your own codebase. This manual gives you the workflow to move faster without losing your judgment. The core thesis: you drive, AI types.
Engineers who ship production software and want to move faster without losing their judgment. Not hobbyists. Not people who want AI to "build my app." This manual assumes you already have opinions about linting, type systems, and test strategy.
| Chapter | Description |
|---|---|
| Introduction | The core problem -- five ways AI erodes engineering agency -- and the mental model to prevent it. |
| Tools | What your AI coding tool needs (plan mode, LSP, forking), OpenCode setup, and MCP servers. |
| Project Setup | Scaffold it yourself, guardrail tooling (linter + type system at max strictness), and writing AGENTS.md. |
| The Loop | The Plan -> Execute -> Test -> Commit workflow that separates agentic coding from vibe coding. |
| Testing | No-mock philosophy -- test real behavior with real dependencies. Package-level test structure. |
| Anti-Patterns | When NOT to use AI, the acceptance spiral, common failure modes, and the three-attempt rule. |
| Appendix: UI With Real Designs | Screenshot-driven and token-driven approaches to turning real designs into code. |
drsh4dow.github.io/agentic-coding-field-manual
bun install --frozen-lockfilebun run lint:mdmdbook buildThe generated site is written to book/.
src/-- Markdown source for all chapters and appendicesbook.toml-- mdBook configuration.github/workflows/-- CI (markdown linting) and GitHub Pages deployment
MIT. See LICENSE.