spruce.world

How to Find a Problem to Work On

advice and observations after a year in the trenches

22.09.25 4:00am

Throughout this last year helping run prod, I met a ton of people who are, like me, interested in changing the world with technology. Most commonly, people want to start a startup. They give lots of reasons for this: most commonly, they're actually lying to themselves about these reasons and it boils down to relatively personal feelings about ambition and impact. It remains the case that there are certain people that are really good at startups, and it is the right answer for these people.

The obvious next question is which startup to start. There is plenty of literature on this topic of “finding a problem” which you may read (eg this YC slop). I don't think it's very helpful. However, I have spent the last year searching for a problem, and watching a bunch of others do it, and have now happily left the process behind. I've concluded that it's all bollocks, and this post will explain why and give some more practical advice.

There's a concept in decision theory called the “explore-exploit tradeoff”. This also comes up in reinforcement learning and game theory and a bunch of other fields: basically, an agent can always either be searching for more solutions (exploring) or executing a known solution (exploiting). Search forever and no progress is made, exploit forever and you may overlook a better path. When I told him I was in search of a problem, my friend Isaak told me to explore more. He explained that if I just took a step back from what I was working on and started exploring, I would find something that truly excites me.

The problem with trying to explore more is that “finding more problems to work on” is not a real activity. You cannot start doing it at 9am, and stop for tea at 5. You cannot work overtime on it and find even more. I have seen incredible people try to “find more problems to work on” for weeks, and achieve absolutely naff all. Invariably, you go back to whatever the first idea was before you started. Everyone who I've spoken to who did this has said it was a waste of time.

In reality, there are only two ways to get new ideas for what to work on:

1.Do Nothing. This is a highly underrated tactic. Some people feel extreme pressure to start working on something, to drop out immediately to do it, and to pick any idea as long as it lets them start. This is stupid - in the best case scenario, you will fail quickly, and end up where you started. In the worst case scenario, you will fail slowly or experience mild success, thus locking you into putting all your energy for years into an idea you don't care about. The point of doing nothing is that in the course of doing a university degree or perhaps an unrelated reading spree, an idea may come across you. You generally won't know immediately that a given idea is the one, but over the following months if it's still stuck in your head, it's a good sign. Unfortunately there is no way to speed up this process or force it to happen; you cannot focus really hard and have good ideas (tangential ai analogy: in my experience reasoning models are far less creative than a few shots of a good base model). For this strategy to work, you have to seriously Do Nothing - work if you must, but only on something random or unrelated. If you're trying, it won't work. The only thing you can really do to speed up Doing Nothing is to make sure your life is full of entropy: keep reading weird shit or going to random places etc

Unfortunately, Doing Nothing doesn't always work - most of the old generation of startup literature on this topic (e.g. this PG essay) suggests focusing on solving problems you experience in your everyday life, but I think we've outgrown that as a society - in other words, it's been patched. I've written before about how all the problems in college sophomores' lives (food, dating, cabs) have been solved. The problems in e.g. a large corporate finance team's lives are not as obvious. Moreover, if you are not an expert in a problem, it is hard to understand how to solve it, and expertise in these non-obvious problems takes effort. It is not hard to be an expert in web apps; it is hard to be an expert in jet engines.

Becoming an expert in a real-world problem rarely comes from reading a book. If you could read books about the problem, then it would be academic, and these are not the sort of things startups do - you can get funded to do a PhD instead. Talking to people who are already experts is a good plan, but if the existing solutions to the problem are really bad, it's quite likely that they don't actually know anything - remember, if they really had all the answers, they would be doing a startup themselves.

2. Do Something. This is an even better strategy than Doing Nothing, but it's harder to do well. Importantly, this does not need to be the actual idea - please do not attempt to like, move house and hire people and raise funding before you know what you're doing. But it should be in the right area, and it should teach you something about the system you're trying to change. A good way to do this is to find the best team that's currently working on something related and join them. This is what I did to find out more about healthcare - I joined OpenAI's Health team, where I've had plenty of time to talk to doctors and hospitals and all sorts of people who wouldn't normally meet, and try out ideas I wouldn't normally be able to. This is a solid plan because if the team is really great, you'll learn a lot by osmosis, avoid falling into the expert trap I mentioned above, and make some friends along the way. A example of how to do this badly is my friend Max, who tried to learn more about healthcare by getting a job as a receptionist at a doctor's office, and will admit himself that it was a waste of time - he got to see a lot of how the health system looks, but was not able to effect change or test new hypotheses about it, found no interesting people, and got no meaningful artefacts or leverage by doing it. Doing Something only works when the something you're doing matters - and you cannot kid yourself about this. If whatever you're doing is real, you'll be continually getting feedback and trying new ideas, thus getting new data - you are learning things about the problem that nobody else knows. If you aren't getting new data, you might as well be talking to experts or reading a book. This new data will eventually be so rare that it is a secret - and as we all know, a secret is something you can start a startup on.

So there it is - if you're looking for an idea, either Do Nothing or Do Something. If you don't have one yet, don't feel pressure - if you yourself aren't completely convinced it will work and (crucially) actually right about it, it almost certainly won't. There are very few accidentally great successes, and you shouldn't bet your life on it happening to you.

I'm not sure if I have all the answers with healthcare yet. I may not have found out all the secrets this summer, but nobody does when they start. The process of finding an idea to work on doesn't stop when you have the idea, since nothing comes out fully formed. The best I can hope for is confidence that I'm working on the right thing, and that I know enough to have an edge over everybody else. I think I'm nearly there.