Skip to content

Agentic Software Engineering: Build, Verify, Evaluate

Agents generate. You verify, decide and explain.

Over twelve teaching weeks and an assessment week you learn to deliver changes in a real codebase with a coding agent, confirm that the agent got them right, and explain and modify your own work without an agent. The course is taught in English.

Read the syllabus How this course works Start with Week 0

How this course works

  • Generate, then verify

    Every example shows both steps. A check you can rerun is evidence; an agent saying "done" is a claim.

  • Predict before you look

    Each class opens with a decision you can get wrong. Save your prediction, run a check that tells the answers apart, then revise.

  • Decide, don't regenerate

    When a candidate passes, accept it, narrow the task, recheck something specific or stop.

  • Explain it without the agent

    Two checkpoints and a project defense, 30 of 100 marks, are done without AI.

Three lines of work

Six assignments, mostly in the first six weeks; four contributions to two real open-source repositories; and one term project, alone or in a pair, on a topic you choose.

Weeks Topics
1-4 Agent internals, context engineering, skills, and customizing your agent and repository
5-8 Agent-ready codebases, agentic code review, tests that notice a fault, security
9-12 Background agents, evaluating agents, AI-native teams, the software factory
13 Project expo and individual defenses

Assessment at a glance

Component Marks When
Term project: proposal, technical specification, alpha, final 20 Weeks 2, 4, 7, 12
Project defense, AI-off 15 Week 13
Four open-source contributions 30 Weeks 4, 8, 9, 12
Six assignments 15 Weeks 1-3, 5, 6, 10
CP1: AI-off diagnosis, test and repair 7.5 Week 6
CP2: AI-off code review and recovery reasoning 7.5 Week 11
Participation 5 All term

You pass with at least 60 overall and at least 15 of the 30 AI-off marks. The syllabus and the schedule say what is due each week.

The calendar