Agent Design, Context Management and Evaluation
A hands-on session on when an agent is the right solution, how context and oversight decisions are made, and how performance is measured.
- For
- Product, operations, and innovation leaders; technical teams
- Duration
- 4 hours
- Deliverable
- An assessed example agent flow and an evaluation framework
- Format
- Online or in-person
An AI agent is a system that decides on steps towards a goal, uses tools, and judges its own results. This program treats agent design through practice rather than vocabulary. Participants examine when an agent is the appropriate solution and when it adds complexity for no return, then work through the relationships between planning, memory, context selection, human oversight, security, and measurement. Throughout the session, tool definitions are written and multi-step flows are built on enterprise AI tooling, so the group can see where a flow breaks and why.
What the session covers
- The core concepts and terminology of agents, and the criteria for choosing the approach at all
- How an agent breaks a goal into steps, how it recovers when the plan fails, and when it should stop
- Memory and context management: what to give the agent, what to leave out, and what those choices cost
- The decision, approval, and intervention points where a human has to stay in the loop
- Common failure modes, the security risks that come with tool access, and the controls that answer them
- Building tool definitions and step-by-step flows to observe the breaking points, then evaluating performance on task completion, error rate, latency, and cost
Participants leave able to justify an agent approach, design the context and oversight decisions behind it, and assess results against consistent measures.