A few things I’ve been thinking about: Note #2
As a refresher, I am testing sending these occasional notes to some people in my network to share a few ideas I have been thinking through. This one is about a question I keep returning to:
What happens to user interfaces when agents start doing more of the work for us?
A couple of years ago, I worked on a report about generative UI. The idea was exciting, but mostly impractical, and early. Models were too slow, too expensive, and too unreliable to generate useful interfaces in real time.
A lot has changed since then. A lot has not.
Claude now shows artifacts inside the chat when a visual or interactive surface works better than text. Vercel released its JSON-render GenUI framework. More teams are experimenting with interfaces that are generated around the task instead of fixed in advance.
And yet, most software still looks much the same as it has for the last 20 years: tabs, dashboards, menus, and generic screens.
The broader question I am asking myself is: What should software look like when users no longer need to click through it the same way?
One answer is that the UI disappears.
As agent capabilities improve, more work will happen in the background. Instead of opening five tools, searching across tabs, we will give an agent a goal, give it access to the right tools, and let it work. In that world, the interface becomes mostly headless. The user might only see a an approval request, or a final deliverable.
For many workflows, that is probably right.
But I do not think all software becomes invisible.
Sometimes the agent only needs a yes or no. Other times, humans need to inspect, compare, correct, decide, or understand. And when the question is complex, text alone is often a poor interface. A chart, table, timeline, map, dashboard, simulator, or custom control can help someone understand faster and make a better decision.
That is where GenUI becomes interesting. Not UI for the sake of UI. The point is to make software easier to use, more useful, and better matched to the task. I also think it has potential to solve one of the biggest challenges many apps face: guiding the user towards the art of the possible.
Most software today is designed for the average user. The same screen has to work for everyone, which means it is rarely perfect for anyone.
AI makes a more adaptive personalized interface possible: One simple version I keep wanting is a single tab I open every morning. It would show what matters most to me: the agents waiting for my approval across apps, my priorities for the day, the people and companies I should follow up with, the meetings that matter, and the score of my favorite soccer team. Likely there is a 60%-70% that is fairly standard each day, but I believe the real magic is in the other 30%: the part that changes because the system understands what matters today in a way that works for me.
The challenge is that dynamic interfaces are hard.
If the UI is slow, users will not use it. If it misunderstands intent, the experience gets worse. If it breaks brand consistency, companies lose trust. It also increases the chance of runtime errors, broken screens, and strange edge cases.
From the investor and founder lens there is also a deeper question: where does the value accrue, and who owns the interface?
Does every company build its own adaptive UI? Will Anthropic and OAI own this, or is there room for a middle layer that understands user preferences, and with a simple line you give it to your different agents and apps.
Either way, the current model of 30 tabs, rigid dashboards, and generic screens feels overdue for a reset.
A few companies I have been tracking in this space include Thesys, CopilotKit, UI MCP, and SkyValley. I was excited on the application side for what Dreamer was doing before they got acquired by meta. Most of the first wave still feels early.
If you have seen a magical experience around UI, or want to discuss adaptive software, I would love to hear about it.
And as always, if you or someone you know is thinking about raising or starting a company, even very early, and wants a friendly investor perspective, I would love to connect.
Till next time,
Daniel