AI Ledger: Git-native governance for agentic AI
An append-only standard for declaring intent, keeping scope explicit, and recording what AI agents actually changed.
AI agents make it easy to change more than you intended. That is useful when the direction is right and dangerous when assumptions, scope, or side effects remain invisible.
AI Ledger is a Git-native standard for AI change traceability. Its core idea is simple: governance should live beside the work, use the tools developers already understand, and preserve history instead of rewriting it.
Declare intent before meaningful changes
Before an agent begins substantial work, the ledger records the intended outcome, constraints, scope, and success criteria. This creates a contract that both the human and agent can inspect.
After the work, the ledger records what actually happened. The difference between the plan and the result becomes visible instead of disappearing into chat history.
- Declare intent before meaningful changes
- Keep scope explicit
- Record what actually happened
- Preserve an immutable, append-only history
Governance without a second bureaucracy
The standard is designed to fit Git rather than compete with it. Records can live on a separate branch or worktree, keeping governance history available without cluttering the product changes themselves.
The initial v0.1 release included a CLI that could be tried with:
npx @ai-ledger/cliThe project is open source at github.com/ai-ledger/ai-ledger.
What dogfooding revealed
Later dogfooding suggested something more interesting than compliance. Invisible self-governance improved the agent's own work.
Writing down intent, constraints, and success criteria before implementation reduced drift. Append-only reflection after the change created a better record of decisions. In my own early use, the improvement felt substantial, roughly from 80% to 95% alignment, though that is an anecdotal estimate rather than a benchmark.
The most promising lesson is that traceability does not need to slow an agent down. When the process is lightweight and native to the repository, clearer intent can improve both oversight and execution.
Adapted from a post originally published on LinkedIn.