Building Dillio has given me a front-row seat to what AI does exceptionally well and where it still falls short.
The technology is remarkable.
It can generate ideas, write code, create designs, and help explore solutions at a speed that would have been hard to imagine just a year ago.
But after spending months building a real product with it, I’ve noticed three areas where it consistently struggles.
1. It doesn’t truly understand people.
It can analyze data and identify patterns, but it doesn’t understand what it’s like to stand on a pickleball court, run a league, organize an event, or deal with frustrated players when something goes wrong. I think this is one of the hardest areas to identify AI hallucinations.
2. It’s incredibly agreeable.
Ask AI for feedback on an idea and there’s a good chance it will find a way to support it. That’s not just my experience. Researchers studying large language models, including teams at Microsoft, have identified “sycophancy” as a common behavior, where AI tends to agree with users even when it shouldn’t.
3. Maybe most importantly, it doesn’t know what not to build.
Coming up with features is easy. The difficult part is deciding which ones don’t belong. Most products struggle because they try to do too much, not because they do too little. What’s interesting is how closely these three limitations are connected. Understanding people helps you question assumptions. Questioning assumptions helps you avoid building things nobody actually needs.
Over the next few weeks, I’ll expand on these points and share some examples from building Dillio and where we’ve seen each of these limitations show up in practice.
The technology is evolving quickly. The fundamentals of building useful products are much more durable.