I Stopped Writing the Code. That Was the Hard Part.

For twelve years, writing the code was not something I did. It was who I was. Then I bet the agency on AI, and the person who wrote every line had to step aside.

The reason I made that bet is duller than the announcement makes it sound. I am a solopreneur. AI is everywhere. Staying on top of the game meant betting on it, and I would rather make that call while I still had a say in how it went than have it made for me a year later.

A one-person agency has no bench. Whatever the market decides is normal to expect, I either deliver it myself at that speed or I find another way to deliver it. When the thing the whole market is recalibrating around is AI, sitting it out is a decision too, and it is the expensive one. So I made the call, and then I lived with what it cost.

The strategy-deck version says we repositioned Essential Code around AI. That is true, and it leaves out the only part that actually mattered to me.

What I was really giving up

When I ask myself what was hardest to let go of, the answer comes down to one thing:

Writing all code by myself. Being in the midst of it. My identity as a software developer.

I built TRIGO for twelve years and left in 2023 to build things myself. Both chapters had the same through-line. I was the one in the work. Hands on the keyboard, in the middle of the problem, the person the hard bits ran through. That was the story I told myself about who I am, the identity underneath the work.

Betting the agency on AI meant that story no longer held.

I stopped selling myself as the maker

I am not going to walk you through how any of it works under the hood. That is not the point, and it is not what this site is for. The business version is simple. I stopped selling myself as the maker.

For years the thing a client bought, whether they knew it or not, was me in the midst of it. Now the thing I sell is the outcome, and I am no longer the one typing every line to get there. On paper that is a smart move. There is clearly a market for it. Off paper it took a while to sit right, because I had quietly built my whole sense of worth on being the maker, the person whose hands were actually on the work.

The part that actually hurt

Being in the midst of it was the part I loved. That never makes it into a repositioning announcement, and it is the entire reason this one was hard.

The middle, specifically. The hour where the shape of a problem finally clicks and you are the one who made it click. Stepping out of that felt, for a real stretch, like losing the thing that made me me. It happened slowly and quietly, more like watching a habit you are proud of stop being yours.

That is what the bet cost me, paid in something no invoice ever shows. I am not going to pretend I made peace with it in a weekend. What I did not see at the time is that the same attachment was quietly costing me something an invoice does show.

The call I got wrong

Ask me what I have got wrong since making the bet and I do not have to think about it:

Waiting too long for an autonomous setup to do stuff by themselves.

I moved too slowly. I could see it coming. What I could not do was trust a system to run without me sitting in the middle of it, because trusting it meant admitting the middle did not need me. The attachment that made letting go hard is the same attachment that made me slow. Those were the same problem wearing two faces. The hesitation had a price, and I paid it in months.

For a solopreneur that hesitation is expensive in a particular way. There is nobody else to go and build the thing while you keep your hands on the client work. If I do not set it up, it does not get set up. So every month I spent waiting to feel ready was a month the setup did not exist, and I was still standing in the position I had already decided to stop standing in.

If you are a founder circling the same decision, that is the part I would flag. The readiness to hand work off usually arrives before you are ready to stop being indispensable. The gap between those two is where the cost lives.

Two things other founders say that I think are false

Two claims come up over and over when I talk to other agency founders about AI. I think both of them are wrong, and since I have bet my own business on this, I am not going to repeat either one to sell you something. They are:

That it can do x. And that they can 100% identify AI content.

“It can just do x”

Take the first one. Somebody says AI can just do x, whatever x happens to be that week, and the room nods, because everyone in it has seen the demo where it does x. The demo is real. The demo is also one run, on a clean input, with nobody’s name on the output.

Delivery is the hundredth run, on whatever the client actually has, with my name on it. Everything the demo skips lives in that gap: getting the input into a shape the thing can work with, defining what “good” means precisely enough that a machine can hit it, checking what comes back, and owning it when the check fails. Adding AI moves that work somewhere else. It becomes setup, context, and review, and somebody still has to do it.

The other half of the gap is that the same tool hands two people wildly different results, because most of the quality sits in the setup around the model rather than in the model itself. One person gets the demo result and the next gets mush, same week, same tool. So when a founder tells me AI can do x, what they usually mean is that it did x for them once. That is a real data point, and it is a long way from something you can sell as a service and deliver every Tuesday.

“I can always tell when it is AI”

The second claim is quieter and more sure of itself. Somebody tells me they can spot AI content every time, that nothing gets past them.

Two problems with that. The first is that you only ever count the ones you caught. Whatever went past you does not put its hand up afterwards, so the confidence gets built out of hits with no idea how big the pile of misses is. Anyone can feel accurate under those conditions, including me.

The second is that every giveaway anyone can name to me is already on somebody’s list of things to strip out before publishing. So what is being spotted is mostly unedited output, which is a much smaller claim than the one being made in the room.

And under both of them sits a question that rarely gets asked. Made by AI to what degree? A person decided what it should say. A person cut the parts that were wrong. If you can still point at the sentence that gives it away after all of that, you are better at this than I am. I work with these tools every day, and I would not stake my name on being right every time. That is the part that bothers me. The people most certain they can spot it tend to be the ones handling it least.

You would think betting the agency on AI would make me its loudest fan. It made me its most careful skeptic. When you actually depend on a thing, you stop kidding yourself about its edges fast, because those edges are now your problem. I sell the outcome, and I stay honest about how it gets made. I will not tell you a machine did something it did not do.

So who am I now?

Still a builder. The altitude is what changed.

I used to build the thing. Now I build the system that builds the thing, set the direction, and make the calls about what is worth making at all. It turns out the part I thought was my whole identity, hands in the code, was always a means to an end. The end was the outcome for the person paying, and that was true long before AI was in the room. I just could not see it while my sense of worth was tied to the typing.

This is a shift I earned by getting it wrong first, slowly, and in public.

I stopped being the person who writes the code. I did not stop being a builder. That took me longer to believe than it should have, and it is the truest thing I know about this whole bet.

Yours, David