0/35FDE path lessons understood. Go to your dashboard.

 

Topic 03 of 10

Agents & tools

Agents that change things in someone else's systems: the loop, tools, approval gates, failure handling, and when a plain workflow is the better build.

  • 12 lessons
  • 7.6 h reading
  • 12.8 h hands-on
  • 6 on the FDE path
Start lesson 01
Progress0 / 12

Lessons

  1. 01The agent loop, written by handPath · stage 4A while loop around a model call, with state and stop conditions you can name. Pro
  2. 02Agent or workflow? Choosing the autonomy levelPath · stage 4Most use cases want a fixed workflow with one model call. Know when they don't. Pro
  3. 03Designing tools a model can use wellPath · stage 4Names, descriptions and schemas decide whether the agent picks the right tool. Pro
  4. 04MCP: connecting agents to the customer's systemsServers, clients and transports, and what to check before you trust one. Pro
  5. 05Tools with scoped, auditable accessLeast-privilege credentials per tool, so the security review says yes. Pro
  6. 06Memory that makes people come backWorking, episodic and semantic memory: what to store, what to forget, what to protect. Pro
  7. 07Workflows that survive restartsDurable execution for processes that run for days through outages and deploys. Pro
  8. 08Approval gates and bounded autonomyPath · stage 4Grant autonomy step by step, with people approving the actions that matter. Pro
  9. 09Retries, fallbacks and failing gracefullyPath · stage 4 LabDesign for the model being wrong, slow or down, and keep the work moving. Pro
  10. 10Subagents, skills and code modeSpecialisationSplit the work without multiplying latency and tokens. Pro
  11. 11Rules, SQL, models, solvers or an LLM?Path · stage 1Pick the cheapest technique that solves each decision, and use the LLM where only it can.Free
  12. 12Coding agents in someone else's codebaseSpecialisationUse coding agents to change a legacy system safely: characterise first, then migrate in slices. Pro