About this course
Study software systems in which a language model acts through tools, alongside the practice of building software with coding agents. The course follows one repository and two agents: the agent a team builds, and the coding agent used to write that code under tests, logs, and gates.
Twelve Sunday classes form four parts. Every class pairs a one-hour lecture with a two-hour lab, an in-room pass check, and a measurement in the results ledger. Each part strengthens the same repository and ends with a deliverable that the next part builds on.
Current stage: Part 1
I'm currently taking Part 1 — The Model as a Measured Component, covering Classes 1–3. Parts 2–4 are the upcoming course plan. These pages contain the supplied syllabus and lab briefs; individual attendance, study dates, completed labs, and measured results have not yet been recorded.
The source is Agentic Software Architecture: the course direction, a draft dated 4 November 2026. This site uses the owner's course name, Agentic Software Engineering 2026. The draft schedules classes from 8 November 2026 to 24 January 2027, with the final paper due 31 January 2027. Calendar, staffing, assessment, and budget decisions remain provisional.
Download the original 65-page course direction.
Learning outcomes
By the end of the course, students should be able to:
- Estimate the cost, latency, and reliability of an agent from model and token properties.
- Design an agent from tools, retrieval, and memory, with contract tests and recorded architectural decisions.
- Deliver software with coding agents under specifications, locked tests, architecture rules, and security controls.
- Evaluate an agent architecture with repeated trials, calibrated graders, cost, and time.
- Formulate, run, and report a reproducible experiment on agentic systems.
Four-part roadmap
| Part | Planned dates | Focus | Part output | My stage |
|---|---|---|---|---|
| 1 | 8–22 Nov 2026 | The Model as a Measured Component — cost, latency, reliability, wrappers, schemas, agent loops | Tested three-tool agent, pass^3, run records, part1-ref | In progress |
| 2 | 29 Nov–13 Dec 2026 | The Agent as Tested Subsystems — retrieval, MCP, context, memory | Agent + MCP + C4 ADR, part2-ref; due 19 Dec | Upcoming |
| 3 | 20 Dec 2026–3 Jan 2027 | The Harness as a Control Layer — configuration checks, specifications, test locks, trust boundaries | Hardened agent and coding-agent controls, zero-bypass probe results, part3-ref | Upcoming |
| 4 | 10–24 Jan 2027 | The Architecture as an Evaluated Claim — calibrated graders, comparisons, peer reproduction | Four-page paper, peer reproduction, decision memo; final paper due 31 Jan | Upcoming |
Part 1: start here
- Class 1: Agent anatomy, reliability and token cost → LM wrapper and tokenizer lab.
- Class 2: Inference capacity and structured output → Schema contracts and extraction lab.
- Class 3: Tested agent loop and evaluation → Three-tool agent and pass^3 lab.
Keep personal measurements in each lab's results ledger. Scheduled dates are not study-completion dates.
Lectures
Class and lab schedule
| Class | Scheduled Sunday | Lecture and paired lab | Ledger measurement |
|---|---|---|---|
| 1 | 8 Nov 2026 | Agent anatomy, reliability, token cost · Lab 01 | Bangla/English token-cost ratio |
| 2 | 15 Nov 2026 | Inference and structured output contracts · Lab 02 | Invalid rate, field accuracy, cost, spread |
| 3 | 22 Nov 2026 | Tested agent loop · Lab 03 | pass^3 per task and run records |
| 4 | 29 Nov 2026 | Retrieval as a tested subsystem · Lab 04 | Recall@10; pass^3 before/after |
| 5 | 6 Dec 2026 | Tools and MCP contracts · Lab 05 | Contract tests and injection audit |
| 6 | 13 Dec 2026 | Topology, context and memory · Lab 06 | Success and tokens per arm |
| 7 | 20 Dec 2026 | Silent harness faults · Lab 07 | Found/seeded faults; skill trigger accuracy |
| 8 | 27 Dec 2026 | Specifications, locked tests, layer rules · Lab 08 | Hidden-test result, tokens, cost, minutes |
| 9 | 3 Jan 2027 | Agent trust boundaries · Lab 09 | Bypasses on both agents |
| 10 | 10 Jan 2027 | Calibrated graders and trajectories · Lab 10 | Kappa; pass^k with interval |
| 11 | 17 Jan 2027 | Harness architecture comparisons · Lab 11 | Cost and minutes per accepted task |
| 12 | 24 Jan 2027 | Reproduction, validity and thesis paths · Lab 12 | Reproduced headline number |
Key takeaways
[Personal takeaways from Part 1]