Coaching Agents for Training, Language, and Growth
Four narrow coaching agents over my own data: a strength-training coach, an English coach, a PM coach, and a personal-growth coach.
Why
The agent ecosystem proved that a narrow agent with deep context beats a broad one with shallow context. I wanted to test the same idea on myself, where the data is small, personal, and honest.
The four coaches
Kinetic coach. A read-only connection to my training journal. Eight modes: reflect on a session, find weaknesses, check progression, weekly digest, plan a mesocycle, decide whether to deload, audit exercise variety, and check goals. It reads sets, volume, sleep, and readiness signals and argues from the numbers.
English coach. A structured learner profile, curriculum paths, error-pattern tracking, vocabulary review, and writing feedback, all in the vocabulary of my actual work: steering committees, vendor calls, and regulatory language.
PM coach. Retrospectives on my own sessions, missed opportunities mined from meeting transcripts, grounded improvement exercises, and progress tracked over time.
Personal coach. Shadow work, habits, and inner-life goals. Deliberately partitioned from the PM coach so career growth and personal growth never blur.
The partition is the design. A coach that can talk about anything ends up saying nothing specific. Four coaches with hard boundaries each say something I can act on.
What I learned
Personal data forces honesty in the retrieval layer. If the coach hallucinates a personal record, I notice immediately. That feedback loop made these agents the best test bed for the grounding and self-critique protocols the whole ecosystem now uses.