The first thing I published about The Wheel, back in February, was about the floor rising with the access AI can give you. I’d spent four and a half months building an encryption system with an LLM as my development partner, and my argument was that AI lets one person get into fields that used to take a decade or a team to enter. When enough people do that, the ceiling moves too, because the good ideas tend to come from connecting fields. I was pretty sure the people who’d want The Wheel would be people working that way: using AI to think further than they could alone, not to do their thinking for them.
When we pictured our first customer, it was an AI-native company. Those are the companies signing up for a new tool every week, with people pasting work into whatever assistant gets them through the day, often without anyone signing off on it. That’s what people have started calling shadow AI, and it’s the problem I wrote about in June: the record of the work scattering across tools nobody chose on purpose.
We were getting ready to raise money, and a pitch works best with a tightly defined customer. The numbers were there. Working from Census business data, we counted about 163,000 US startups under five years old, with 50 or fewer people, already using three or more AI tools. Those companies have budgets, lawyers to review things and maybe even early procurement processes, which is what an investor wants to see. And our early conversations with founders had gone well.
Those founders were worried, and they aren’t alone. In a UK survey that Startups.co.uk ran with Sage this July, 90% of AI users said they were concerned that what they enter could end up training someone else’s model. Half of the founders surveyed had shared sensitive information with public AI tools in the past month anyway. That matched what we heard. As the conversations went on, we found that while a lot of founders think The Wheel could be useful, many have already either accepted some level of risk or found a way around it that works for them. The Wheel is for them too. Some of them may just need more time.
Since we announced the beta, I’ve been talking with more and more people about how they actually want to work with AI, and the conversations that go somewhere are increasingly happening outside the startup world. Some are artists who want help with grant applications, which means handing over a statement of practice and the half-formed idea for the next piece. Others are mathematicians who want to work through unfinished problems and pull in the papers they lean on, or policy researchers testing an argument against interviews they’ve promised to protect.
Most of them haven’t signed up for Instinct or Muse. Plenty haven’t heard of either, or they’ve seen the furry creature on Muse’s subway ads (if you’ve ridden the subway in New York lately, you have too) and are still trying to figure out what it can actually do for them.
How much AI they use varies a lot. The mathematicians are already deep in it. In September I wrote about two who spent most of a year on one problem with AI and then had to ask a company whether their drafts had gone into its model. Others are just getting started, pasting their documents into chats with a tinge of regret and curiosity. Stories like the mathematicians’ keep coming, and the din has gotten too loud to tune out. What doesn’t vary is the work itself. It’s their thinking, and they want AI to push it further without standing in for it.
Thinking with AI needs somewhere to keep the work
In February I called continuity the missing piece. Good work doesn’t happen in one conversation. It happens over weeks, with dead ends that shape the next attempt and connections that only show up after you’ve wandered through three neighboring fields. When every conversation starts from scratch, you spend your time re-explaining the problem instead of working on it.
What carries over, the drafts and dead ends and half-made connections, is the most private part of your work. A grant application lays out a piece of work before it exists, and a proof in progress is the same. I wrote then that your thinking is yours until you choose to share it, or until enough other people figure out the same thing that it becomes part of the floor. While you’re still working something out, though, it shouldn’t be sitting readable on someone else’s servers or feeding someone else’s model. The artists, mathematicians and researchers I’ve been talking to already know that, and a lot of them have decided the safest place for it is nowhere near AI.
Start with your work, not with AI
Mostly, we made it work for people who aren’t already living and breathing AI: people who don’t know what a skill is and have no interest in learning to optimize a prompt. We went through it screen by screen and took the jargon out.
The first thing you do is bring one of your own documents and ask it a real question. There’s a short tour afterward if you want it, and you can replay it from Help. Studio, which everyone has, is the part of The Wheel for making things instead of just asking about them. That’s where the grant application gets drafted, from your own notes and in your own voice, and what Studio learns about how you write stays sealed in your Personal workspace.
Receipts are written in plain language. Each one says what went in, what came out and whether anything left, and when The Wheel uses its own AI or searches your files, it says that in those words. Your browser checks that nobody edited the receipt afterward. A shared space is called a workspace, not a team, because a lot of shared work happens in groups that aren’t teams, like a working group spread across five organizations. Each workspace is walled off from the others, and nobody needs an admin to set one up.
Underneath, notes and files are encrypted in your browser before they reach our servers, and they’re searched blindly. Most of the AI work runs on small models we host ourselves, which don’t log what they’re given or what they produce. Outside AI stays off until you turn it on, it only goes to providers that keep no data, and The Wheel strips names and identifying details before anything leaves. Nothing you put in trains a model.
It still works for companies
A company is a group of people doing exactly this kind of work, and The Wheel is the same at every size: your Personal workspace, a workspace for three people, a workspace for a whole company, with the same encryption and the same receipts. The first hour is designed for the person doing the work, not the admin rolling it out, and a company gets the same product.
A big company can handle this with an enterprise contract or by running AI on its own servers. A small AI-native company might try to do the same, build its own tools, or decide the risk is worth it to stay on the newest models. Everyone else, the three-person studio or the researcher working alone, gets the standard terms. So a lot of them have kept their real work out of AI.
Before it becomes the floor
I ended that first post by saying the floor rises for everyone and the ceiling is still yours to reach. The floor rises because people share what they figure out, on their own schedule. The Wheel is for the part before that, while the thinking is still yours, or your group’s. It’s also where you decide when and how it goes out: into a shared workspace for the people you work with, or to the other AI tools you already use, through a connection you set up yourself. Anything that goes out through it comes with a receipt, and you can switch the connection off whenever you want. If that sounds like your work, come talk to us.