Beyond
"It Works"
A practical workflow for building features with AI coding agents
First edition · 21 September 2026
26 pages · 8 prompts · 5 templates · 1 worked example
Take one small feature from a vague request to a reviewable change, with evidence that it does what you asked.
PDF · 146.2 kB · Free to read. No signup.
From "done" to something you can check.
For developers and solo builders who already use a coding agent and know their way around a repository, a diff, and a running application.
The agent can implement the feature and run checks. You define the outcome, set boundaries, evaluate the evidence, and decide what is acceptable.
The method stays tool-agnostic, with short, sourced notes for Codex and Claude Code.
- Define what done means
- Give the agent useful context
- Bound the plan, then build
- Recover without random rewrites
- Review evidence, then the change
- Leave a usable handoff
Teaching preview: make the request checkable.
- Vague request
- “Add CSV export.”
- Concrete requirement
- Export all four filtered Open issues, including matches on other pages—not just the two rows currently visible.
Keep the useful pieces close.
Start with the quick-start page. Choose a small feature, write acceptance examples, and use only the prompts and templates that help.
- Eight copy-ready prompts
- Clarify, inspect, plan, implement, diagnose, check, review, and hand off.
- Five editable templates
- A task brief, repository guidance, acceptance checklist, completion report, and session handoff.
- One local worked example
- Add CSV export to a filtered issue table. Includes source, tests, a baseline, and a deliberately faulty teaching variant.
ZIP · 274.3 kB. Includes the PDF, manuscript, prompts, templates, sources, verification notes, and local demo. Browser settings may open a file instead of downloading it.
A local example, with clear limits.
The six issues and deliberate failure are synthetic teaching material. Checks cover this local example; they do not certify production readiness or universal spreadsheet safety.
This guide was created with AI assistance in research, writing, implementation, and production. The example’s implementation checks were run by an AI coding assistant.