Understanding Systems or Users?

Living through a technological disruption is humbling. The AI field is moving forward at such an incredible pace, that yesterday’s principle is today’s fallacy. One of the ways this manifests itself is how different approaches compete for a single solution. Let me demonstrate.

Last week, I wrote about how applications are increasingly built for LLMs rather than humans. The idea is that humans want natural interfaces such as text, voice, touch, maybe even smell? The objective is to solve a task - not to sit behind a computer and click around.

LLMs enable us to get closer to these modalities by translating information between machines and humans. However, this necessitates that LLMs understand the machines - which is not always trivial. That’s where the idea of designing for LLMs instead of humans comes in.

However, there is another approach to solving a job-to-be-done. Instead of designing systems for LLMs, which is time-consuming as you have to repurpose your entire IT-landscape, you could imitate the user. Major research labs are taking this route. They work on products which enable artificial agents to navigate UIs designed for humans. It’s an interesting approach. The LLM doesn't have to understand the systems - it has to understand you (and the websites you like).

Which idea wins is to be seen. The tools to imitate users are already here. In addition, I don’t expect organizations to radically adopt building for AI. So it's easy to expect user-imitation to prevail. However, don’t underestimate the constraints legacy systems (i.e. systems designed for humans) might impose on the future value of organizations. You’d be surprised how many companies have been disrupted by a trend they didn’t see coming.