I agree. If everyone on my team is just prompting agents, why would I spend money on their salary when I can use that for tokens instead? No one has yet given a good answer for this question when I've asked before.
Because agents alone cannot create good software. You are paying for people who understand how to use the agents. LLMs are a tool, and software engineering as a discipline ain't dead yet.
I already understand how to use them as a professional engineer myself, so I don't see the benefit especially for the amount of money a salary and benefits cost.
Then maybe you don't need them. Only you can answer that for yourself. That's kind of the point of my post. It seems like you're looking for an argument here, and I don't want to argue. I have no idea whether or not you need a team. Maybe you'd be better off spending money on tokens. I don't know.
I'm not looking for an argument, this is a discussion forum so if you think it's an argument, then that's fine, you do you, but my point is that the cons you listed don't actually seem like cons in this day and age with AI so I am questioning and validating assumptions anyone would have as well; for example my friend whom I talked to recently is also a solo technical founder and he too doesn't think it's worth it to hire people anymore.
Counter question: where's the useful production-grade software that's been made entirely with "agent swarms"?
Personally, I don't think useful business problems can be solved entirely with LLM loops. If you choose not to pay others to handle part of that then it's all on you.
Is it worth paying 200k a year for that, especially if I'm already good enough at prompting? That amount gets me a lot of Claude Code licenses instead.
Depending on scale of your products you may find that single person won't be able to cover all aspects. "Prompting" requires some domain knowledge and also quality control and guidance.