The Idea That Wasn't One
For the better part of a year I went looking for the idea that would justify leaving my job. The most useful thing I did in that time was build a Raspberry Pi cluster I was certain would never matter.
For the better part of a year near the end of my time at Goldman, I thought my real job, the one that would actually decide my future, was to find the idea. This was the work underneath the work, the thing I did in the evenings and on weekends and quietly told myself was the point.
Not just any idea. The idea. The kind an investor would recognise immediately, the kind that justified walking away from seven years of a career I was already preparing to leave. I carried a notebook everywhere. I filled it with half-formed businesses I never built and never will.
Most of what I wrote in that notebook was rubbish, but I did not know that yet. I just knew it did not feel like anything.
I remember sitting in a co-working space in London with my friend Nkechi. She was in the middle of leaving banking for software, teaching herself by building small things, little projects with no reason to exist except that building them taught her the work. I was doing the opposite. I was staring at a spreadsheet of "market opportunities" I had built to convince myself I was making progress. I was not making progress. I was performing the shape of it.
Watching her build for the sake of building unsettled something in me. She was not waiting for permission, or for the perfect idea, or for a career to reorganise itself around her first. She just wanted to be good at it, so she built her way there. It reminded me, a little uncomfortably, that I used to be like that too. That I actually liked this, and would be doing it whether or not anyone was paying me to.
At some point I closed the spreadsheet and said something like: I don't want to think about billion-dollar ideas today. I want to build a Kubernetes cluster.
It was not a strategic statement. It was the opposite of one. Kubernetes had nothing to do with fintech, nothing to do with Africa, nothing to do with anything my next company was supposed to be about. I just found it interesting, and after months of trying to want the right things, wanting the wrong thing felt like relief.
So I bought a stack of Raspberry Pis and built a cluster.
It went badly at first, in the ordinary way these things go badly. Nodes that would not join. Networking I did not understand well enough to debug, so I had to go and actually learn it. Not just enough to make the error go away, but enough to know what was happening underneath. I set up ZeroTier so I could reach the cluster from anywhere without opening my home network to the internet. I put a Cloudflare tunnel in front of it so I could expose a service without ever exposing the box itself. Then, because I had gone that far already, I 3D-printed a case for it. There was no reason to. It just felt like the kind of thing you finish properly or not at all.
My notebook was full of businesses that had never made a single mistake, because not one of them had ever been built. The cluster made mistakes constantly. That was the whole difference.
None of this was going to be a business. I was fairly sure of that at the time, and I was right. That was never the point.
The skills, though, were real. Networking that does not fall over. Infrastructure you can trust because you built it yourself and know exactly where the weak points are. The instinct to expose a service safely rather than carelessly, which turns out to matter enormously when the thing you are eventually moving across borders is other people's money.
I wasn't building for what came next when I made that cluster. It didn't exist yet. But the muscle I built that year was a specific one: the comfort with infrastructure, the refusal to ship something whose internals I did not understand. It is the same muscle I lean on now, building things that have to work in places where the network is the least reliable part of the stack.
The cluster is still running. It lives under my desk, three Raspberry Pi nodes, and I still call it cutiePi. I think of it as the home lab where ideas get proven before they earn a domain, which is a slightly grand way of saying: this is where I try things before I trust them anywhere else. Right now it is running Hermes, an agent orchestrator I built, mostly because I wanted to see what it could do without paying for cloud compute to find out.
I do not think there is a clean lesson here, and I am suspicious of posts that insist there is one. But if I am honest, the most useful thing I did in that whole year of hunting for the idea was the one thing that was never meant to be an idea at all. Not because it produced a business. It did not. It produced a person who was slightly harder to fool with infrastructure that only looks like it works. And maybe with a few other things too.