Things I haven’t built yet

Building iKo has changed my sense of what one person can make. It has also meant spending a fair amount of time keeping up with what’s becoming possible.

Over the past year, I’ve repeatedly discovered that something I was about to build had already been solved. Sometimes it was a service tucked away in Cloudflare. Sometimes a model had quietly learned to do something it couldn’t do a few weeks earlier. Keeping up takes time, but it saves a lot of reinventing.

It also keeps giving me ideas. These are a few things I’m curious about right now (October 2026). Some might become projects; others might disappear from the list when I discover what someone shipped tomorrow.

A home for software nobody wants to maintain

Creating and sharing an app is already straightforward. I’m interested in what happens when a business accumulates a hundred little tools. Could they inherit the right access to company data, remain dependable as systems change, and survive their creator leaving? Having run an innovation lab, I can think of plenty of useful experiments that become worth trying if they’re also affordable to keep alive.

Working together while the agent keeps going

Agents can already join team conversations and work on tasks over several days. What I’d like to explore is how people direct that work together: whose instruction takes precedence, how a changed decision affects tasks already underway, and how someone takes over without reconstructing the history. A shared customer case might be a useful place to start. I’m interested in how responsibility works when several people and an agent are moving the same piece of work forward.

Specialist services delivered by generalists

AI helping less experienced workers is already familiar. The bigger possibility is a business organized around that capability: hiring capable generalists to deliver a service that previously depended on scarce specialists. Someone might handle a complicated benefits application, with AI guiding the investigation and qualified experts reviewing the cases that need them. My interest is in the whole service: what it costs, how quality is checked, and how much more accessible it becomes.

A business that builds its own intelligence

Training and hosting a customized model is already possible. What interests me is making improvement part of everyday operations. A corrected mistake becomes a test; a well-handled case becomes a training example; a model update has to earn its place by doing the work better. Could an ordinary business run that cycle and retain the models it develops? Building iKo has made me particularly interested in that connection between experience and measurable improvement.

A tutor that grows up with a child

AI tutors already personalize their help using a learner’s history. What draws me further is a tutor that develops with a child over years, beginning with reading and arithmetic and gradually helping them reason independently. It would need to judge when to explain, when to let them struggle, and whether understanding lasts. Working on a conversational coach has made me curious about that kind of sustained relationship, and how you would establish that it helps someone learn.

AI designed around older people

I’m interested in what AI could offer older people across everyday life: conversation, learning, practical help, staying connected, or recording memories for their family. Products such as ElliQ already explore parts of this. What draws me is designing something around the person that can adapt as their interests, routines and abilities change. Someone could move naturally from discussing a book to understanding a letter or arranging a visit, without navigating a collection of apps. I’d enjoy exploring how to make that relationship useful, enjoyable and easy to live with.

Less friction in Swiss public services

Switzerland is already experimenting here. Zurich’s AI sandbox developed a prototype with Kloten for preliminary checks on construction notifications. Its findings also exposed practical limits, including inconsistent interpretation of plans. My experiments with Swiss building standards make me interested in the next step: could a system work through a dossier, connect the relevant cantonal and communal requirements, and explain what needs correcting? An applicant could resolve problems earlier, while the commune receives a clearer file and a reasoned assessment to review.

If any of this overlaps with something you’re building, thinking about, or struggling with, give me a shout. I’d enjoy comparing notes.