[fzwang]

I talked to a number of people who used AI to start their business in the early stages.

Some of the positive things I see is that vibe coding is pretty powerful for non-technical people. If you adopt the AI-as-floor-raiser mindset, it removes a major barrier for experimentation for people who don't know how to program. Some go a bit overboard and think they don't need engineers at all, but they eventually realize they still do. People either throw out the vibe coded prototypes or had to rebuilt them "properly". Some technical folks appreciate having this option available to non-tech people, as it shortens the experimentation cycle and acts like a filter to weed out bad ideas. One interesting new dynamic is that we could be seeing a new wave of fast-follower builders, where a good engineering team can replicate vibe-coded products quickly, in a way that's more durable and without reliability issues.

Using LLMs for market research is pretty mixed. The good cases seems to be using it for writing generic routine stuff, ie things that are not core to your business/product, but just need to exist. The bad is somehow expecting ChatGPT to identify novel business ideas. It seems like one of the major benefits of chatbots is actually forcing you to write down your ideas. For example, some of the LLM generated "ideas" are pretty generic and run of the mill stuff you'd find anywhere. But it was interesting to review the prompts and chat history, as a human reviewer, to get a sense of what the person is trying to ask. For example, when I work with students, it's great to see their chat/prompt history to "reveal" their thinking process. It works the same w/ entrepreneurs vetting ideas. It'd be interesting to have a highly hallucinogenic model just as an idea mixer.