Design, build, evaluate, and govern AI agents with the rigour that survives production.
An interactive course that turns the Agent Canvas, the REMIT governance framework, and current evaluation and compliance practice into hands-on artefacts you can take into your job tomorrow.
4 pillars, one workflow.
Each pillar has lessons, diagrams, interactive tools, a glossary, and a quiz. The Scenario Builder ties them together into a ready-to-use agent.
Problem framing, user needs, agent scope, tool selection.
Architecture, prompting, RAG, orchestration, MCP, skills, memory.
Evals, test sets, LLM-as-judge, red-teaming, monitoring.
Risks, oversight, REMIT, NIST RMF, EU AI Act, RSP.
Hands-on, not hand-wave-y.
Fill in the 8-cell Agent Design Canvas and the 5-pillar REMIT worksheet. Save, export, share by URL.
Walk through the six ingredients of a good system prompt, with examples from production agents.
See how model tier, tool calls, and iterations move the cost and quality bands. Built for knee-finding.
Plot cost vs benchmark across frontier and open-weight families. Switch benchmarks to re-rank by use case.
Walk a realistic use case through all four pillars. Outputs a ready-to-use system prompt, Agent Canvas, and REMIT worksheet.
Run a toy eval against a sample agent with an editable LLM-judge rubric. Watch how rubric changes move the scores.
Try to jailbreak a sandboxed demo agent. See which attacks succeed, which fail, and why.
“Find the simplest solution possible, and only increase complexity when needed.”
— the through-line of this course.
Built on the shoulders of giants and robots
Four-pillar course architecture and the Agent Canvas by Dr Mark Bloomfield (Cambridge Judge Business School). Interactive adaptation, additional lesson content, and engineering by Simon Elliston Ball. Much of the writing, diagrams, and code was produced in collaboration with Claude — a fitting choice for a course about building with agents.