I think the issue you're overlooking is: what happens to your product if model providers further enhance their capabilities?

To be honest, I once had a similar idea—to build a system like this. But I gradually realized that model providers are becoming increasingly powerful, and they’re even venturing into application development themselves, not just focusing on the models. This makes anyone building applications on top of LLMs seriously consider whether the model providers might eventually replace them and do a better job.

I can even give you what I think is a very representative example: Manus. You can’t possibly not know this product. But now, both OpenAI and Google are developing their own full-featured agents. If Manus can’t come up with new ideas, I think they might be in serious trouble.