24 lessons

Agentic Engineering

A post-doctoral curriculum from intermediate to expert: harnesses, tool protocols, context and memory, evaluation science, inference economics, and the security and oversight questions that arrive once an agent acts on its own.

A working engineer’s course on building agents that hold up outside a demo. It starts from the harness rather than the prompt — the action space, the observation function, the verifier — and treats the model as one component of a system whose behaviour you have to measure.

The mathematical prelude is not decoration. Evaluation statistics decide whether a result is real; verification theory decides whether a check catches what it claims to; the cost model decides whether a deployment pays for itself. Everything after that leans on those three, and the course returns to them in every module.

Modules run from tool interfaces and structured action, through context engineering and long-horizon state, into harness optimisation, inference engineering, and trajectory fine-tuning. The later third is about what happens at scale: multi-agent architectures, sandboxing and execution environments, agent security, oversight and governance, and production operations. It ends with a cheat sheet — the whole curriculum compressed to the propositions, thresholds and decision rules worth carrying.

  1. Agentic Engineering: A Post-Doctoral Curriculum from Intermediate to Expert Level
  2. MODULE 1 — Build First: The Minimal Harness, Then the Formal Foundations
  3. MODULE 2 — Tool Interfaces, Structured Action, and Protocols
  4. MODULE 3 — Context Engineering
  5. MODULE 4 — Reasoning, Planning, and Test-Time Compute
  6. MODULE 5 — Evaluation Science for Agents
  7. MODULE 6 — Anatomy of a Harness and Component Classification
  8. MODULE 7 — Execution Environments, Sandboxing, and Verification Loops
  9. MODULE 8 — Long-Horizon State, Memory, and Decomposition
  10. MODULE 9 — Architectural Constraints and Entropy Management
  11. MODULE 10 — Automatic Harness Optimisation and Self-Evolution
  12. MODULE 11 — Observability, Cost, and Latency Engineering
  13. MODULE 12 — Open-Weight Landscape and Selection Methodology
  14. MODULE 13 — Inference Engineering
  15. MODULE 14 — Model–Harness Co-Adaptation and Heterogeneous Fleets
  16. MODULE 15 — Trajectory SFT, Distillation, and Data Engineering
  17. MODULE 16 — Agentic Reinforcement Learning
  18. MODULE 17 — Environment Engineering, Verifier Engineering, and Verification Economics
  19. MODULE 18 — Multi-Agent Architectures and Protocols
  20. MODULE 19 — Agent Security
  21. MODULE 20 — Oversight, Alignment, and Governance
  22. MODULE 21 — Computer-Use, Research, and Domain Agents
  23. MODULE 22 — Production Operations and Organisational Adoption
  24. THE COMPLETE AGENTIC ENGINEERING CHEAT SHEET

Take the whole course