Book · From $9.99
When Agents Fail
A recovery lab for agents that change real things
Your agent says it stopped. What did it actually do? A tool call times out, someone presses Stop, a result arrives after the goal changed, an approval gets used for something it never covered. This book takes those moments one at a time: each chapter names a failure, reproduces it in a small synthetic run, builds the code that handles it, and ends with an exercise you run on your own machine.
Who it’s for
- Engineers building agents that call tools with real side effects: payments, orders, emails, records.
- Tech leads who need to say what a failed or stopped run actually did, and recover without doing it twice.
- Anyone who learns by running code: every chapter ends with an exercise in Python, standard library only.
What you get
- What actually happened Confirmed, failed, unknown: say what a run actually did, keep a run log that survives a crash, and return an honest summary after Stop.
- Not doing it twice Retry without paying twice, reconcile from evidence, and fence out stale workers.
- When the world moved Results that arrive after the goal changed, approvals that expire and recipients that change, and an agent that says it finished.
- Recovering honestly Reverse, compensate, and tell people the truth, then hand a broken run to the next owner.
- Never again Turn every failure into a test, then into a team drill.
- The companion lab, free 16 runnable exercises with checks and reference answers. Python 3.10+, standard library only, no accounts or keys.
Terms
- Formats
- Kindle ebook ($9.99) and 7 × 10 in paperback ($24.99, 304 pages), first edition, November 2026. Sold and delivered by Amazon.
- ISBN
- Paperback: 9798141005718.
- The lab
- MIT licensed, free at timurista.ai/labs/when-agents-fail and on GitHub, whether or not you buy the book.
Honest notes
- Every scenario, shop, product, person, log line and number in the examples is synthetic, written for teaching.
- Independent publication by Tim Urista. Not affiliated with or endorsed by any AI vendor.
- Test recovery code against your own systems before you rely on it.