The Improvement Loop — Agents that get better in production
**Twelve essays on the engineered feedback cycle: what improves, what breaks it, and what bounds it (PDF + EPUB)** Shipping an agent is the start of the job, not the end of it. The uncomfortable question comes after: is it any better in month three than it was on day one — and if it is, what exactly improved? The model didn't retrain. Something else moved. This book is about that something else. Twelve essays on the improvement loop — the engineered feedback cycle that makes deployed agents better with use. Part I draws the frame: proof that a self-improving stack already exists, and the line where automation ends and agency begins. Part II walks the substrates that actually improve — harnesses that build themselves, skills as compound interest, memory you can train, and the world-model turn. Part III is the loops themselves, with their failure modes: training on your own exhaust, the verification ceiling, judges on trial, and the Red Queen dynamic where evaluator and evaluated co-evolve. Part IV goes beneath: the serving loop that turns inference-time compute into capability, and the paradigm bet waiting after autoregression. It is the operational sequel to *Agentic Engineering*: that volume is about building agents that ship; this one is about what happens to them after they ship. The two cross-reference each other, and each stands alone. Every claim was traced to its source before publication: numbers, attributions, and mechanism descriptions audited against the originals, and anything that couldn't be verified marked rather than smoothed over. It is a **living book**: versioned like software, and every future edition is included with your copy. 138 pages · 12 chapters in 4 parts · every figure redrawn as vector · per-chapter references · appendices. **Who it's for:** engineers running agents in production who want “it gets better over time” to be an engineering claim, not a slide. **Who it's NOT for:** readers wanting model-training internals or a no-code tour — if that's you, the purchase will disappoint, and there's a 14-day no-questions refund either way. **What's included:** - `improvement-loop-v2026.07.pdf` — the book, 138 pages - `improvement-loop-v2026.07.epub` — the same book for e-readers - `README.txt` — what's in the purchase, the living-book promise, and the companion volumes - `LICENSE.txt` — personal use, no redistribution The thirteen checklists PDF is **not** in this product — it belongs to *Agentic Engineering* and ships with that book. **Terms:** 14-day no-questions refund · personal-use license; keep it on any of your own devices; no redistribution or resale; brief quotations with attribution are welcome · living-book delivery: every future version (v2026.MM) is included at no cost through Gumroad updates, with corrections listed in Appendix B.
Get it → dattgoswami.gumroad.com