Multi-agent design, MCP and evaluation
The model-agnostic layer: how to decompose a job into tools an agent can use reliably, when a second agent earns its keep, and how to evaluate a system whose output is different every time you run it.
Ordered the way we teach it. Private workshops reorder and reweight this against what your team already knows.
Naming, schemas and error messages an agent can actually recover from — the highest-leverage work in any agent project.
What to load, what to summarise, what to leave on disk, and why a long context window doesn't remove the problem.
Exposing your internal systems through a standard interface, with authentication and permissions that survive an audit.
Orchestrator and worker patterns, handoffs, and the honest cases where one well-scoped agent beats a committee.
Deterministic checks, rubric grading and model-as-judge — plus how to keep the judge itself honest.
Tracing, cost ceilings, failure containment and the human approval boundaries that let you sleep.
Teams often pair a model track with Agentic Systems — the tool design and evaluation material applies whichever provider you settle on.
Send over your stack, your group size and roughly when you'd like it. We'll come back with a curriculum outline and a date.