I've been using GitHub Copilot since it became available in the technical preview in 2021. During this time the field of AI developer tools has changed dramatically. Now we have many alternatives to Copilot such as Cursor, Windsurf and Aider. The tools have not only improved the autocomplete suggestions developers get but also changed the workflow of developers using them. Cursor specifically was the first tool I felt was a major step forward in using AI as coding assistant. It seemed to understand the context and the project I was working on much better than Copilot and was able to make large changes spanning multiple files. Recently GitHub has added to Copilot the support for editing multiple files and providing a larger context for the LLM to work with similarly to what Cursor does. It seems to work reasonably well in most cases even if I have to make some changes to the results to make them usable. I think this is already a huge step forward from the initial Copilot versions.
Steve McConnell famously wrote that managing complexity is the most important technical topic in software development. Complexity in my day to day work currently is largely the result of technical debt. Some of the technical debt might be self inflicted due to some uninformed decisions in the past but most of it comes in the form of so called "legacy" projects. And technical debt can be brutal. In the worst case it can completely stop development of new features before it is properly addressed. And almost always it is demoralizing to the developers working on the project.
So what I really want from AI are tools for managing technical debt so small teams and solo devs can effectively work on larger projects without getting bogged down by some framework upgrade which seemingly has no upsides.