AI coding tools are everywhere right now, and most developers are already using at least one of them. What’s less clear is how they actually differ once you move past marketing pages and start using them on real code, with real constraints, and real consequences.

In this talk, I'll put GitHub Copilot, Claude, and Cursor through the same set of realistic scenarios and compare how they behave. Not just greenfield demos, but inside an existing codebase where context, tradeoffs, and prior decisions actually matter.

We’ll look at how each tool handles common development tasks like understanding unfamiliar code, modifying existing behavior, refactoring, and responding to ambiguous requirements. More importantly, we’ll examine where they tend to help, where they quietly get in the way, and how their failure modes differ in ways that matter to working developers.

The goal isn’t to declare a winner. It’s to give you a clearer mental model of what these tools are good at, what they’re not, and how to use them without over-trusting them. You’ll leave with practical guidance on choosing the right tool for the job, setting better expectations for yourself and your team, and staying firmly in control when AI has the keyboard.