Real stakes. Real curiosity.
Building what couldn’t be built two years ago. Skipping what shouldn’t be built at all.
Explore the workAn applied AI studio, built for problems that used to be impossible.
Until recently, a lot of good ideas died in the planning meeting. The budget wasn’t there, the technology couldn’t do it, or it needed more people than anyone could hire. AI moved many of them within reach. That’s where we work: mostly on our own products, occasionally on someone else’s.
The other half of the job is subtraction. A lot of what companies want automated shouldn’t be done at all: the report nobody reads, the review that never changes a decision, the step everyone does and nobody can explain. Automate those and you get the same waste, faster. So before we build anything, we ask why it exists.
That’s rarely a technical question. It’s about how a team plans, prioritizes and decides what matters, which is why we spend as much time there as we do on the software.
Downstream
The good ideas are downstream.
Most AI thinking stops at what can be automated today. We’re more interested in what happens next, once a capability gets cheap and shows up everywhere. The model is only part of the problem once it starts shaping decisions under pressure.
The most ambitious work comes from seeing how things that look unrelated actually connect, whether they’re technical, financial, legal, managerial or political. This is a sample of what we’re pulling on. It’s all over the place on purpose.
What we do
A moonshot that falls over in week three is just a demo.
Ambition only counts if it survives real users. That means software that holds up under heavy use, keeps working when the model or provider underneath it changes, and assumes the model will sometimes be confidently wrong. Not prompt libraries. Not workflow diagrams repackaged as strategy.
We take on very few things at once and go deep on each, doing the unglamorous engineering behind software that’s still running six months later. The interesting part is rarely in the demo. It’s in the architecture that lasts.
From the lab
Zwischen
Your coding agent can write the code. It still needs someone to question its approach, catch the detours and send it back when “finished” isn’t what you asked for.
Zwischen gives that supervision to Frank, an AI co-driver who follows the work as it happens. Frank and your agent resolve most issues between them, bringing you in when they need your judgment. You can get on with your own work.
Get in touch
What are you thinking about?
Tell us what you’re trying to build, or what you’re not sure needs doing. We can start there.