600 Decision Logs in 2 Months
600 Decision Logs in 2 Months
Most teams struggle to maintain a decision log. Meetings happen, decisions are made, and six months later nobody remembers why the database was chosen or why the API was structured that way.
With AI agents writing code, every commit is a decision log - if you structure it right.
Git Commits as Documentation
When an agent makes a commit, the message can include not just what changed but why. The reasoning that led to the architectural choice. The alternatives that were considered and rejected. The constraints that drove the decision.
At 129,000 commits, this is not just a changelog. It is an institutional memory. Every decision is searchable, timestamped, and linked to the exact code change it produced.
600 Decisions in 60 Days
In a two-month sprint with multiple agents, over 600 non-trivial decisions were recorded:
- Architecture choices - why this pattern over that one
- Dependency selections - why this library and not the alternative
- Performance tradeoffs - what was sacrificed for speed and why
- Security decisions - what was restricted and the threat model behind it
None of these required a separate documentation process. They were embedded in the workflow.
The Compound Value
Decision logs become more valuable over time. A commit message from three months ago explaining why a particular approach was chosen saves hours of re-investigation when someone asks "why is it built this way?"
The cost of writing good commit messages is seconds. The cost of not having them is hours of archaeology through code trying to reconstruct the reasoning. Agents that write detailed commits are agents that pay forward.
Fazm is an open source macOS AI agent. Open source on GitHub.