Thesis: Context Engineering for Autonomous Software Development: Bridging the Model Capability Gap

Smaller language models can lower the cost of autonomous software development, but often perform less reliably than frontier models. This thesis explores whether carefully selected and structured context can narrow that performance gap on realistic coding tasks. It also examines how context relevance and volume affect results, including when extra information becomes counterproductive.

The assignment

• Develop a reusable, model-agnostic context engineering pipeline for autonomous coding agents, supporting local and cloud-based inference. Preferably build the prototype on an existing agent harness such as OpenCode; running smaller models locally is optional.
• Compare persistent project documentation, manually prepared task-specific information, and automatically retrieved or generated context against an agent with no additional prepared context but normal repository exploration tools.
• Evaluate smaller and frontier models on reproducible feature development, bug-fixing and refactoring tasks in representative industrial-style repositories, using automated tests and explicit success criteria.
• Vary the amount of relevant, redundant and unrelated information to identify diminishing returns and potential performance degradation.
• Measure functional task success and changes in the capability gap, alongside execution time, inference costs, context preparation effort and pipeline maintenance complexity.
• Separate the effects of information prepared in advance from information discovered by the agent, keeping tooling, repository states and execution conditions comparable.
• Keep model training, fine-tuning, new agent framework development, extensive hardware benchmarking and local inference optimization outside the project scope.

Who we are looking for

• Master of Science students in Computer Science, Software Engineering or a related discipline.
• Expected graduation in June 2027.
• Fluent written and spoken Swedish and English.

What we offer

• A thesis placement at Knightec Group’s Sundsvall office, within a company working with product and digital service development.

Practical details

Location: Sundsvall office. Start: January or March 2027. Apply through the company’s careers page as soon as possible, by 30 November 2026. Questions can be sent to johanna.edstrom@knightec.se.