Stop asking AI what you should build
I'm seeing more and more people use AI to decide what ideas to work on.
LLMs seem smart enough at lots of hard things, like writing code. Why not talk to them to help figure out what to build?
I think this is a Very Bad Idea for two reasons.
Firstly, the best ideas are really problems people face. Ideally this starts with yourself or a small group where you can see the secret problems which are hidden to outsiders. But for people who aren't in this group, the solution will often feel weird and dumb.
Often times, the initial use-case actually IS tiny and weird. But because it's authentic to you and your users, it will keep evolving into something really special.
This means that, while LLMs excel at giving you the average consensus opinion, you in fact need the opposite. Your best ideas will usually be dismissed by Claude or ChatGPT (along with reasonable human beings / investors / friends).
If LLMs had existed in 2005, would they have told you to build a company helping people rent out air mattresses for conferences?
There is a particular flavour of this where LLMs are also really scared of competition. They are often meek and will tell you not to build something because lots of people are building it. Whereas the reality is that a crowded space is often a terrible reason not to build a product [1].
Secondly, I think something profoundly damaging happens when a founder outsources consensus. This could be to an investor, an LLM, or some rote wisdom which everyone else believes. The minute you stop taking responsibility for what you’re building, you make shitty decisions.
This applies not only to the idea itself, but even how you should build it, how it feels, and how you take it to market. Great products evolve organically and creatively (where your loss function is contact with users) instead of being intelligently designed.
So what should you do? Stop outsourcing and figure it out for yourself by talking to your users. And if you are doing that already and you're stuck, you should go and talk to more of them. [2]
One of the reasons is that the world is often much more complicated than you think. You don’t know what you don’t know. This means you probably don’t even know the question to ask the LLM in the first instance. So you're stumbling around in the dark with a flashlight, seeing the world only through a tiny aperture.
The best conversations I have with users leave me with answers to new questions I wouldn’t have thought of asking an LLM. I am often completely blinkered to some angle of the product I haven’t considered. My priors are usually wildly incorrect. And the only antidote to this is listening.
Sometimes I wonder if it's actually a better idea to work on ideas which LLMs think are stupid.
[1] a) you’ll probably build something unique once you work with users; b) other people are often worse than you; c) you are more likely just to die if you triangulate yourself off of other companies and optimize for the uniqueness dimension vs. what your users are telling you. [2] Lots of founders talk to users but are not talking to enough. How do you think talking to 2/3/5x as many people in the next 30 days would affect your trajectory? This could also be trying to sell your product.