Has anyone successfully used AI agents for software development projects?

I recently used AI to help migrate from styled components, and it was surprisingly efficient, cutting down on what would have been a lengthy process. It completed the task in about 35 minutes! However, I found it essential to set clear guardrails; otherwise, the agents tended to drift away from the original task.

Yes, I've had some success with AI agents, but it did take some trial and error before I got it right. Setting up clear context files and guidelines helped keep the agents focused and on track. Without that framework, things can go haywire really fast.

I think what you're describing sounds a bit like vibe coding. From what I know, the closest we've seen to AI-driven software development is mainly using AI to migrate tests between frameworks. There's been some success there, but I wouldn't say it's fully autonomous software development as you might be imagining.

I see the use of AI agents as more experimental than anything. It’s like having a bunch of apes trying to work together without supervision. Sure, some people claim success, but I'm not convinced it's effective beyond basic tasks or in fully autonomous roles.

I've had my fair share of frustrations with AI in projects. It often feels like it's more hassle than help, but there are some who find it useful. I think treating AI agents like junior developers is key. If you provide clear guidelines and context, they can really come in handy.