The AI boom is exciting, there’s so much opportunity and so much to build at the application layer. But the question is: what to build?
Two theses I’ve seen for AI application startups:
- Accept fate, get lucky: The big foundation model providers will build everything and startups will be crushed, so go for short-lived ‘flash in the pan’ apps or aim to be acquired.
- Focus on blind spots: Solve research/engineering problems the labs aren’t focused and aim for technical defensibility.
On the first one, I don’t consider luck a strategy. This could work, but as a founder I’d have a hard time building something that isn’t sustainable or without a long-term vision.
On the second, I don’t disagree, but it needs more specificity. If it’s something the labs aren’t focused on, it’s also likely something that’s structurally unstable.
For example, it’ll get addressed or fundamentally change as the labs make progress and AI evolves (see the startups that get wiped out every model update). Not sustainable.
Therefore, the imperative is to focus on things labs can’t focus on, rather than aren’t focused on. Not blind spots, but blind constraints.
The low-hanging fruit example of this is things that work across models, where the foundational model companies are less incentivized to build. Cursor is a great example of this.
There is so much opportunity and the barrier to building has gotten so low, it’s just such an amazing time to be alive. The hardest question now really is what to build. Find the blind constraints.
