Death of the IDE?

I think if you are approaching the idea of programming by coming from a place of "we're going to spin up a fleet of agents", you've lost the thread entirely. If you want to use an LLM and have it work with you on your code, fine. But the idea that the future workflow is going to be a complete abstraction away from the idea of looking at the code at all is madness.

We have no idea what these things actually cost. Right now LLM agents run at hundreds of dollars a month and are heavily subsidized. The actual price to run them at a profit could be 2x what it is now. Could be 4x. Nobody knows! We're all just vibing in a VC-funded fog and pretending this is a sustainable situation.

My suspicion is that in 24 months we'll know the actual cost of these tools and they won't be something every developer has access to. They'll be expensive enough that you'll need to make a business case, scope it, then invoke them on a specific project with a definite goal. Most of your day to day work will be with much dumber tools that effectively serve as autocompletes.

Am I the only one who gets nervous by this whole agent stuff? I've seen them burn through thousands of tokens and producing absolute crap. Not to say that they're not useful, but I think they're overused. Maybe I'm just old fashioned, but I often find that the agents really make some stupid decisions and they happen so fast that a human can't keep up. So maybe move fast and break things isn't the best way to do them? How about plan well and avoid breaking things? Until such time that the agents are not stochastic parrots, I'm going to be very careful in using them.