Self-Coaching Loop
My AI watches my calls and tells me what I keep doing wrong, by name, every time I do it again.
Feedback on how I communicate used to come once a quarter, from memory. I wanted a few real moments from each call that show me how I actually did.
- The same plumbing as the clip pipeline picks up every external call. Only moments where I am speaking count.
- Claude picks 0-5 strong or needs-work moments, each with one skill tag and a blunt note: what happened, the quote, what to do instead.
- A second pass matches each moment to a named habit in a running registry, either still doing it or caught it, and merges near-duplicates.
- Known habits go into the next call's prompt, so repeats get called out by name.
- Clips go to a private folder and a private Slack digest.
- A habit ledger dashboard rebuilds after each run, with no model involved: flags per call against the prior window.
62calls reviewed, May to Oct
8named habits in the registry
27times I caught myself instead
- I can see how I actually work on calls, whether I'm improving, and what to focus on next.
- I get feedback on every call, not once a quarter.
- Habits are named and counted across 62 calls, so I can tell whether they are going away.