I've been starting to do it for the last months, pivoting my company (reindent.com) into an AI-run company, which would probably take a couple of years to get where I envision it.

From what I've learned, you cannot do it 100% autonomously at the moment. AI Agents are very good at one specific task, but for strategic decision making I think we are not there yet. I've tried multiple approaches: AI Agent SaaS (most are very primitive and not standardized), coding my own AI Agents (baded on LangChain) and using OpenAI's Operator. The latter has been the best, most impressive so far.

As for code generation, Windsurf and Cursor suffer from LLM hallucination, each prompt must be carefully crafted so that each modification is atomic.