Workshop agenda
Drawn from the author's next book on agent architecture, in progress
Production AI agents: a two-session workshop for your engineering team
Taught by Igor Bobriakov, author of Production-Ready AI Agents. Live online, two 60-minute sessions a week apart, up to 15 engineers, cameras optional. Your team brings its architecture questions; no code or traces are shared, and the workshop teaches rather than reviews your system.
Before session 1: a short form for your stack (framework, model provider, tracing tool) and three to five questions from your team.
Session 1 (60 min)
| Time | Block | What the team works through |
|---|---|---|
| 0:00 | Where agents break | The agent loop as states, steps and tools, and the failures that only show up in production |
| 0:10 | Traces you can debug from | What to record per step and per tool call, and how to read a bad run |
| 0:35 | Evals before release | How to build a small regression set from your own failures, checked the way the output's consumer reads it, so you catch a prompt change before it quietly breaks a path that worked |
Between sessions (about 30 min, on your own): score your own agent on the ten-question self-check from the book's open lab. Session 2 starts from your lowest rows (send us the row numbers and scores only).
Session 2 (60 min, a week later)
| Time | Block | What the team works through |
|---|---|---|
| 0:00 | Retries that can't repeat a real action | A retry table for every tool that changes something outside the agent: one idempotency key per side effect, and an approval step for anything irreversible |
| 0:25 | Your questions | Your team's pre-workshop questions and lowest self-check rows, answered as general patterns your team can apply to its own architecture |
| 0:50 | What to change on Monday | One change, one owner, one date |
Your team keeps: the slides, a loop map, a retry table, the self-check to score your agent again in 30 days, and the recording of both sessions. Both sessions are recorded for your team's internal use only; we delete our copy whenever you ask, and in any case after 30 days.
Background reading (free): the open lab for the book, a ten-lesson course with a guide for running it as a team (github.com/ai-builders-group/build-production-ai-agents).