Proactive interfaces

Recent tech interfaces, particularly llms, are reactive: you prompt them, they do something for you. Emerging agentic interfaces involve (or will involve) proactivity. They guess actions that the user might be interested in and suggests them or carries them out behind the scenes. Proactivity is where agents go from interns to assistants, where they get agency: the ability to see the action space, decide a course of action, and execute.

But proactivity fails hard when it's wrong. We forgive computers less than humans. A bad cursor suggestion or autocomplete drives me insane enough to turn off both features. Proactive emails begging me to rejoin some service, worse. Even great suggestions can grate—like your mom telling you to wear a jacket when it's cold. I think humans like to think for themselves.

Chris Pedregal had a good analogy: great proactive tech is when you reach for the coffee machine and the coffee is already brewing. Not coffee thrown at you. The interface knows what you want, reduces friction, but doesn't shrink your action space or make assumptions that could be wrong.

For me, this means interfaces that exist on screen and change dynamically, not notifications. Things I reach towards, not things thrown at me. I want proactive agents that coach me in different parts of my life. One that continuously ingests my health, habits, mood, academic notes and data—and suggests interventions or areas of focus. I'm building this over the next couple weeks. Will write a follow-up if it's useful (or surprisingly useless).