AI Agents & Builders
Design, build, evaluate and ship agents that act on your behalf without wrecking anything.
You can already call a model. This course is about systems that plan, use tools and run for many steps: how agents work, how to specify and guard them, how coding agents and multi-agent frameworks fit in, how to evaluate them properly, and how to get them into production with the security and observability they need.
What you will be able to do
- Describe the agent loop and choose the right level of autonomy for a task
- Write a one-page agent specification with tools, memory, budgets and stop conditions
- Work productively with coding agents such as Claude Code, Codex and Cursor, including repository conventions
- Choose between single-agent, orchestrator-worker and framework-based designs, and justify the choice
- Build an evaluation set and regression suite so changes are measured, not guessed
- Harden and ship an agent: injection defences, least privilege, tracing, cost per task and human-in-the-loop design
Course outline
What an agent is
3 lessons · quizThe loop that turns a model into an agent, the autonomy dial, and the four parts every agent is made of.
- The agent loop: plan, act, observe 10 min
- Levels of autonomy and where humans stay in the loop 9 min
- Anatomy of an agent: model, tools, memory, guardrails 9 min
- Module quiz 5 questions
Designing agent tasks
5 lessons · quizSpecify before you build: a definition of done the agent can check, tools it can actually use, memory that does not rot, and limits that stop it.
- Task decomposition and success criteria 10 min
- Writing tool descriptions the model can use 9 min
- Memory: short-term, long-term, and files as memory 10 min
- Guardrails, budgets and stop conditions 10 min
- Exercise: specify an agent on one page exercise · 30 min
- Module quiz 6 questions
Coding agents in practice
4 lessons · quizThe tools professionals use daily, a working rhythm that keeps you in control, and the repository files that make agents behave.
- Claude Code, Codex and Cursor: what they are 9 min
- Working with a coding agent: brief, plan, review, test 10 min
- Repository conventions: CLAUDE.md and AGENTS.md 9 min
- Exercise: ship a small change with a coding agent exercise · 40 min
- Module quiz 5 questions
Multi-agent systems
3 lessons · quizWhen one agent is not enough: orchestrators and workers, sub-agents and hand-offs, running in parallel, and picking a framework without regret.
- Orchestrator and workers 10 min
- Sub-agents, hand-offs and parallelism 10 min
- Frameworks: Anthropic Agent SDK, OpenAI Agents SDK, LangGraph, CrewAI: how to choose 11 min
- Module quiz 5 questions
Evaluation
4 lessons · quizReplace 'it seems to work' with numbers: build a test set, grade with rubrics and model judges, and run it on every change.
- Why vibes fail: building a test set 10 min
- LLM-as-judge and rubric scoring 10 min
- Regression testing prompts and agents 10 min
- Exercise: a 20-case eval set exercise · 45 min
- Module quiz 5 questions
Model strategy
3 lessons · quizWhich lever to pull (prompting, retrieval, fine-tuning), what open-weight models are for and how to run them, and how to route work between models on cost and latency.
- Prompting vs RAG vs fine-tuning: the decision 10 min
- Open-weight models: Llama, Qwen, gpt-oss, Mistral, and running them (Ollama, Groq, vLLM) 11 min
- Cost, latency and routing between models 10 min
- Module quiz 5 questions
Security for agents
3 lessons · quizInjection and exfiltration when the model can act, the sandbox and permissions that contain it, and the audit trail that proves what happened.
- Prompt injection and data exfiltration in agentic systems 11 min
- Sandboxing, least privilege and secrets 10 min
- Auditing and monitoring agent actions 9 min
- Module quiz 5 questions
Shipping to production
4 lessons · quizSee what the agent is doing, keep the right humans at the right points, roll out with a safety net, and keep measuring after launch.
- Observability: traces, tokens and cost per task 10 min
- Human-in-the-loop design 9 min
- Rollout, feedback loops and continuous evaluation 10 min
- Capstone: production readiness review exercise · 60 min
- Module quiz 5 questions