Most software isn't limited by AI anymore.
Every few months, we get a new model that's faster, cheaper, and better than the last one.
But I've come to realize that the AI itself is not the most important bottleneck anymore.
It's the ecosystem of things around it that gets more difficult – designing a workflow that feels intuitive, building trust, reducing latency and making sure that the product feels like something the person will love to use every day.
A good model attracts users once. A good product makes them return.
And that's why I think that the future generation of AI products will not be determined by their benchmarks. They will be determined by reliability, UX and how much friction they cut off the workflow of the user.
curious whether other people who develop AI products share the same opinion or think that model quality is still the most important thing.
Replies
I think you're right. We've reached a point where a lot of cheaper models (like DeepSeek or some other open source models) are good enough for a lot of day-to-day work, and those kinds of models will only become more capable in the coming years. I think now price, development speed, and maybe convenience are probably more important.
Personally, I'm less concerned with how advanced a model is, and more concerned about how many things I can have it doing at once.