A small, open specification for disclosing how humans and generative AI worked together to create software.
AIDM gives repositories a simple README badge such as AI Development: Collaborative, backed by clear definitions and an optional machine-readable profile. It intentionally does not measure how many lines were “written by AI,” assign a quality score, or rank one workflow above another.
Social Contribution
AI Development Mode is an open project created as part of Astzweig's Social Contribution efforts, with the goal of making AI-assisted software development easier to understand and discuss transparently.
Specification version: 0.1.1 (draft)
Status: Open for early adoption and feedback
| Mode | Short meaning |
|---|---|
| Autonomous | AI carries out substantially the whole development process after receiving the goal/specification. |
| Supervised | AI independently owns substantial multi-step work; humans review milestones and redirect as needed. |
| Collaborative | Human and AI shape decisions and implementation together in short, iterative loops. |
| Directed | Human defines requirements, architecture, and approach; AI executes bounded tasks. |
| Assistive | Human performs the core development work; AI supports with suggestions, debugging, tests, docs, etc. |
| Incidental | AI use is occasional and local, such as autocomplete, explanations, or small isolated edits. |
| None | No generative AI was intentionally used in the declared scope. |
| Mixed | Distinct, materially different workflows are common; the project supplies a profile or scoped declarations. |
The modes are ordered by AI agency, from most to least. They describe workflow and agency, not software quality, developer skill, trustworthiness, or the percentage of AI-generated code.
Replace Collaborative with the mode that best describes your project:
[](https://aidevmode.org/)You can also copy the bundled SVG badges from badges/.
For more detail, add .aidm.json to your repository:
{
"spec_version": "0.1.1",
"mode": "collaborative",
"scope": {
"type": "repository"
},
"profile": {
"planning": "shared",
"architecture": "human-led",
"implementation": "shared",
"testing": "shared",
"review": "human-led",
"human_review_frequency": "continuous",
"autonomous_scope": "task"
},
"last_reviewed": "2026-09-16"
}See schema/aidm.schema.json, ADOPTION.md, and the examples/ directory.
Classify the typical workflow for substantive changes in the declared scope. Do not classify based on an exceptional one-off task. If two adjacent modes fit equally well, choose the mode that reflects greater AI autonomy and explain the nuance in .aidm.json. If materially different workflows are common, use Mixed or separate scoped declarations instead of averaging them.
An accurate declaration tells contributors what workflow and review to expect, lets adopters calibrate their own review, and replaces speculation about AI use with a statement the project controls. No mode is ranked or colored as better, so there is nothing to gain by understating AI involvement.
AIDM is not an AI detector, authorship tracker, compliance certification, security guarantee, quality mark, or moral judgment. It does not attempt to infer provenance from source code. AIDM is a self-declared transparency convention.
AIDM practices what it describes. This repository is declared Collaborative; see .aidm.json for the profile.
SPECIFICATION.md— normative specificationADOPTION.md— copy/paste adoption guideFAQ.md— common questionsschema/aidm.schema.json— JSON Schemaexamples/— example declarations and badge snippetsdocs/— GitHub Pages static websiteCONTRIBUTING.md— how to propose changesGOVERNANCE.md— lightweight governance/versioning
AIDM, including the specification, examples, badges, and website, is released under the MIT License so it can be reused widely.