My last quarter in the big job, I went a full month without opening an editor. Not once. I noticed because it felt normal.
That is the part that finally bothered me. For eighteen years I built platforms: payments handling billions in volume, data infrastructure under a marketplace with hundreds of millions of users, ML systems serving recommendations at a scale that still surprises me when I say it out loud. Real work, all of it. But somewhere along the way the work became the work of other people, and my job became making theirs possible. Decks. Roadmaps. Headcount cases. The quarterly ritual of explaining to one set of people why another set needed money. I was good at it. I was also four layers of abstraction away from anything that compiled.
There is a version of this essay that frames leaving as bravery. That version is a flattering edit. I left for a reason closer to restlessness.
The distance from the work
A director’s calendar is a strange artifact. Mine was a wall of thirty-minute blocks, and almost none of them were about building. They were about alignment, the word large organizations use for the friction of having too many smart people pointed at slightly different things. I spent my days reducing that friction. Genuinely valuable. Also not why I got into this.
The teams I ran were good. That made it worse, not better. A well-run org doesn’t need its director in the code, and if you’re in the code you’re probably failing at the job you were hired for. So you back away. You review architecture instead of writing it. You ask the question that exposes the flaw instead of fixing it.
Here is the part nobody tells you about senior leadership. The skills that get you there are not the skills that keep you happy once you arrive. I optimized for scope for a decade, and scope, it turns out, is mostly distance. More people, more systems, more surface area, a thinner connection to any single piece of it. I had traded depth for breadth so gradually I never noticed the rate.
The thing the tools changed
I would not have left in 2021. One person could not do very much then, and the gap between what I could build alone and what a funded team could build was too wide to cross.
That gap closed. Not all the way, but enough. By the middle of this year a single engineer with good judgment, a coding assistant that actually holds context, and a few model endpoints can stand up things that used to need a pod and two quarters. The tools don’t replace teams. The floor moved, right when I was sitting in a job whose entire premise was that output only comes from coordinating many people.
So the bet was simple to state and uncomfortable to make. If a small group with deep context and modern tools now moves faster than a large group with shallow context and process, then the thing I knew best, building coordination machinery for hundreds of engineers, was depreciating. Not worthless. Depreciating. I would rather be early to that than explain it in a postmortem.
The discomfort, named plainly
I am not going to pretend this felt clean. Three things got worse the day I left, and saying otherwise would insult anyone weighing the same move.
The income. A director’s compensation is not a number you walk away from lightly, and the founder version of it, for a long while, is a spreadsheet that only goes down. I count cloud spend like my own money, because it is.
The status. There is a real, slightly embarrassing grief in going from a title people recognize to explaining for the tenth time that no, it is just me right now. Hundreds of people used to be downstream of my decisions. Now my decisions are downstream of whether the deploy worked.
And the blank repo. This one surprised me most. I had forgotten how exposed the start of something feels, when there is nothing between you and the empty file but whether you can actually still do this. I could, mostly. Eighteen years of reviewing other people’s code leaves you with strong opinions and slightly rusty hands. The opinions came back faster than the hands.
What the platform years actually left me
People assume the move from leadership back to building is a step down in seniority. I think it is the opposite. It took doing it to see why.
Eighteen years of platform work did not just teach me how to build things. It taught me which things are worth building, the more expensive lesson and the one you cannot get from a tutorial. I have watched enough systems get strangled by their own cleverness to have a physical reaction to premature abstraction now. I have shipped enough features under real regulators to know that the boring constraints (audit, residency, who is allowed to see what) are the architecture, not a layer you add later. I have run finance for infrastructure, so I cannot build a thing that loops over a model endpoint without an instinct asking what it costs at scale.
That is what I brought to the blank repo. Not faster typing. Judgment about where the bodies are buried, paid for by burying some of them.
The venture is an enterprise AI platform, and the engineering belongs in a different post. This is the smaller, more honest one. I left a good seat at a good company because I missed the work, the tools finally made one person dangerous again, and I would rather build something I can put my hands on than manage the distance from things other people build.
Ask me in two years whether the exchange rate was fair. Today I can tell you only that I open the editor every day, and it still feels like the right kind of hard.