A hand-written manifesto for letting AI write all your code
One developer's field notes from 8+ hours a day of agentic coding — rules, skills, MCPs, and how to stop writing code yourself.
What it does
This is not software — it’s a guidebook. The author, a formerly AI-skeptical coder, documents how they transitioned to a fully agentic workflow where the AI writes every line and the human acts as a “leader/manager” who translates intent into prompts. The guide covers environment setup (rules, skills, MCPs), avoiding AI slop, and even physiology-based habits for sustaining long working days.
The interesting bit
The author insists the entire guide was written by hand with zero AI prompting — a deliberate stance that AI can’t express personal experience. That’s an unusual position in a guide about delegating everything to AI, and it’s honestly the most credible part: the advice reads like lived experience, not generated filler.
Key highlights
- Core philosophy: you become a translator/manager, not a coder — the author hasn’t written a line of code manually since adopting this workflow
- Emphasizes building your own rules, skills, and MCPs rather than downloading them, since generic ones often degrade AI output
- Includes a companion MCP,
project-progress, that keeps a documented roadmap so the AI doesn’t drift when you branch between features - Advocates heavy documentation of features and bug fixes so the AI can reorient without rereading entire files
- Covers prompt refinement as a discipline — review every prompt and rewrite it before sending
Caveats
- The productivity claims (“10x–20x work power”, “any project in 6 months”, “16+ hour days”) are personal anecdotes, not evidence — take them as enthusiasm, not benchmarks
- The guide is tied to the author’s specific setup (Cursor, Claude/ChatGPT/Grok) and their own workflow; they explicitly say their rules wouldn’t fit your usage
- The README itself acknowledges the author doesn’t claim to have the best prompting approach
Verdict
Worth a skim if you’re moving toward agentic coding and want a practitioner’s perspective on rules, skills, and MCP hygiene. If you want measured claims or rigorous methodology, this is a diary, not a study.
Frequently asked
- What is xop01/ai_goodpractice?
- One developer's field notes from 8+ hours a day of agentic coding — rules, skills, MCPs, and how to stop writing code yourself.
- Is ai_goodpractice open source?
- Yes — xop01/ai_goodpractice is an open-source project tracked on heatdrop.
- How popular is ai_goodpractice?
- xop01/ai_goodpractice has 599 stars on GitHub.
- Where can I find ai_goodpractice?
- xop01/ai_goodpractice is on GitHub at https://github.com/xop01/ai_goodpractice.