Are we building IDEs for engineers anymore?

Cursor 3 came out - it looks cool, lots of nifty agentic features - but something kinda burns in the back of my head. Who are the audience for these things?

I was originally going to write about the two new visions for software development interfaces, one, like codex where it defers all agency to your agent. They let you open your code in an IDE, but really the goal is to maximize your agent to do all the work, even leaving comments on code instead of letting you edit the diffs themselves. Then there was the Cursor 2/Zed/IntelliJ paradigm where you as the developer have the option to choose if your agent will help you for tasks. Originally I thought both were sort of missing the point of agentic development, but I think they would eventually converge into something practical for an average developer.

But I think this new Cursor 3 seems to have fallen for the Codex FOMO, and is starting to just mirror the paradigm of an solely-agentic powered builder. Agency as a developer will soon be secondary to your dependence on an LLM. Modes like the visual editor on the surface are really cool; but any experienced developer could probably make the same edits in the faster time, and for cheaper.

This is the ultimate goal for agent based software companies, as your revenue comes from asking your agent to do things for you, not if the user is doing it themselves. Cursor 3, you can’t use without signing in anymore (or at least, it’s pretty hard to get around the login screen).

That brings us to the question; who are we building these IDEs for? If it’s for someone who isn’t technical, this is great; this gives people who want to build, a new tool. But for me as an engineer adding more and more layers of abstraction from the code, especially since models aren’t perfect, seems like a strategy in making things that were once simple, expensive.