For AI coding agents, context plays a central role in shaping output quality, consistency, behaviour and style. This session explores context as an engineering concern rather than a prompting detail.

We’ll look at how agents interpret different forms of context, including how their accuracy can be improved when coding against both open-source dependencies, and internal libraries. We'll look at how context, as a specification, can change the behaviour of a coding agent and steer it's decisions, including testing strategies, and project structures. Through practical examples, we’ll show how context influences code generation as well as other important areas, such as code review, from individual PR-level guidance to repo-wide testing.

We’ll also look at what happens when we overwhelm an agent with context and how this will decay, and reduce reliability, and how to overcome this. By the end of the session, you’ll understand how to structure and maintain context so AI agents produce results that align more closely with your code, architecture, and team practices.

AI coding agents are increasingly part of day-to-day development, but it's easy to misuse them, simply but not using their features effectively. In this session we’ll give hands on advice on how to use agent context, when to use reasoning depth levels, concrete techniques for managing agent behavior, through context structuring, compaction strategies, sub-agent usage, commands, and hooks.

We’ll share how popular agents, Cursor, Claude Code, Gemini CLI, and Codex use these techniques, highlighting meaningful differences in strengths, and limitations. By the end of the session, you’ll have a clearer mental model of how these agents behave, when they are effective, and real takeaways you'll use every day in your agentic development workflows.