Being unsure is a natural property of fast decision models. Alejandro Martinez explains how this uncertainty is actually useful for triggering human-in-the-loop decisions: go.peterfriese.dev/ai-using-je...
alejandromp.comUsing Jev to Help a Human Decide | Alejandro M. P.Something we don’t speak enough about as we include more AI in our processes is the cost, not of the tokens, but of the classical compute we need. The more AI, the more PRs, the more CI we run. To combat that I have been building a custom system I call the merge gate. The gate watches the team’s pull requests and when one looks ready, the gate syncs it with the main branch, which triggers the real CI jobs. It picks the order so CI time stays under control, and then it merges the PR once everything is green. This fully automated process has kept the costs of our CI under control allowing us to increase the usage of AI in a responsible manner.
0 reposts3 likes