AI coding tools are getting very good at producing output. The hard part is knowing what to spend, what to trust, and what the repository should remember after the task is over.
I started Cairn because that invisible tax was becoming impossible to ignore. The same work was being repeated across agents. Model choice was guesswork. Credits disappeared into context and retries. A green-looking answer could still leave no durable evidence behind.
Cairn is my attempt to build the layer beneath that work: a local record of what happened, why an intelligence path was chosen, what it cost, and whether the result held up.
More verified progress per unit of intelligence.
The first version is deliberately close to the tools people already use. Codex and Claude can keep doing the work. Cairn makes the economics and the proof visible. The next layer is GitHub-native: evidence that follows issues, pull requests, reviews, and merges.
If you are shipping software with AI and feel the invisible tax, I’d like to hear what your receipt is missing.