Description
Agent demos are easy. Agents you can leave running are not. This day is about everything between the two: controlling what an agent may do, knowing whether it is right, and being able to debug it at 3am. Written for people who will own the thing in production.
What you will leave with
- Design an agent architecture that fails safely under load
- Implement approval gates and human-in-the-loop for consequential actions
- Build an evaluation set so you can tell whether a change helped
- Instrument agents for tracing, cost attribution and debugging
- Choose sensibly between single agent, multi-agent and plain workflow
Agenda
| Anatomy of an agent: perceive, reason, act — and where each fails | |
| Tool design: idempotency, permissions and blast radius | |
| Orchestration patterns — routing, delegation, and knowing when not to | |
| Guardrails: input filtering, output validation, action allowlists | |
| Human-in-the-loop: approval gates that people will actually use | |
| Evaluation — building a test set for a non-deterministic system | |
| Observability: tracing, cost attribution, failure forensics | |
| Capstone: harden a fragile agent until it can be trusted |
Who it is for
Engineers and technical leads who will be responsible for an agent once it is live.
Prerequisites
Confident Python or JavaScript, and prior exposure to LLM APIs or agent frameworks.
Format
- Duration: Full day (09:00–17:00)
- Level: Advanced
- Group size: maximum 10 attendees
- Delivery: live online, or in person by arrangement
Need this for a whole team, or extended to full length on your own systems? We deliver every workshop as a bespoke on-site course too — talk to us.
Reviews
There are no reviews yet.