Agentic Software Engineering

Introduces the engineering of software with and for LLM-based coding agents, and can be taken in both the bachelor’s and the master’s programme. The course covers how coding agents work, how to build and extend them, how to delegate work to them and verify what they produce, and how to assess and sustain the software that results. It replaces Advanced Topics in Software Engineering, which is discontinued after the 2026–2027 academic year.

A. Foundations of LLM-based software engineering

1. Language models and coding agents: model architecture and training, the agent loop
2. Prompt and context engineering for coding tasks: prompting, context management, compaction

B. The architecture of a coding agent

1. Tool integration: tools, the Model Context Protocol (MCP), API calling
2. Extensibility and harness design: skills, the anatomy of a harness
3. Efficiency and state: caching, long context, cost and model selection, memory and retrieval

C. The software engineering process with coding agents

1. Specification and delegation: specifying tasks, delegable work, managing agents
2. Verification of agent-generated code: review, testing, security analysis, human factors
3. Safety and security of coding agents: threat models, sandboxing, permissions, safety benchmarks

D. Assurance and sustainability

1. Evaluation and monitoring of coding agents: benchmarks, statistical significance, tracing
2. Sustainability of agent-built software: technical, cognitive and intent debt, design for change