Say how AI helped build your software.
AI Development Mode gives repositories a simple, human-readable way to disclose how much agency humans and generative AI had during development, without fake percentages or line-by-line attribution.
One README badge is enough to adopt it. The helper below finds the right one.
Read the specificationView on GitHub
A real example: this project's own declaration. The specification and site were built with an AI assistant in short iterations, with the maintainer setting direction and reviewing every change.
Interactive helper
Find your mode
Answer a few yes/no questions about the typical workflow for substantive changes. You get a mode and a badge snippet to paste into your README.
Reference
The modes
A mode describes the typical workflow for substantive changes. It is not a quality score and no mode is better than another. Modes are listed from most to least AI autonomy.
- Autonomous
- AI performs substantially the complete development process after the initial goal or specification.
- Supervised
- AI independently handles substantial multi-step work while humans review milestones and redirect.
- Collaborative
- Human and AI shape decisions and implementation together through frequent feedback loops.
- Directed
- The human defines requirements, architecture, and approach; AI executes bounded tasks.
- Assistive
- The human performs core development; AI helps with debugging, tests, docs, review, and suggestions.
- Incidental
- AI is only used occasionally for small local tasks.
- None
- AI is intentionally not used at all.
- Mixed
- Materially different workflows are common. Use Mixed with a profile, or publish scoped declarations, instead of averaging them into a meaningless number.
Background
Why declare it
Not a score
Modes describe how work was organized, not whether the software is good, safe, or trustworthy. No mode is ranked or colored as better.
Not authorship tracking
AIDM describes the development process. It does not claim which individual lines came from a model.
Honesty is the cheap option
An accurate declaration tells contributors what to expect and lets adopters calibrate their review. Since no mode looks better than another, there is no reason to downplay AI use.
More detail when it matters
An optional .aidm.json profile describes planning, architecture, implementation, testing, review, and the largest scope AI handles alone.
About
Open by design
AIDM is vendor-neutral, MIT licensed, and created as part of Astzweig's Social Contribution efforts. The goal is a small convention that communities can improve and reuse freely. Contribute on GitHub.