I write less code than I used to. A lot less.
My agents write most of it now. They find bugs I wouldn't have noticed, open PRs, and give me a daily standup with breakdowns of what went wrong and why. I review, I merge, I move on.
I did not expect it to feel empty.
Not because the work got worse. The work is fine. Better than fine. I shipped Kimbundu and PulseEvent this year while holding down a full-time job. Before LLMs that was not happening. Now the bottleneck isn't writing the code. The bottleneck is me: my taste, my judgment, the questions I know to ask and the ones I don't.
But there's a weird grief in watching the thing you trained for become abundant. Code used to be the hard part. Now it's the easy part. The time to MVP has dropped to maybe a month, no matter the project. The barrier collapsed. Building something stopped feeling like an achievement.
The stuff I built for myself
So I started building for myself. If code is cheap, why not take care of myself first?
I want to make a fighting game? beatdown.ambacelar.com is right there. I bought some assets off itch.io and now I'm seeing if I can get a fighting game engine built with good netcode. If it's fun, maybe I'll commission bespoke assets and make a proper game out of it.
I want to play a card game? jack7.ambacelar.com is live, on iOS, and anyone can jump in.
I want to learn Kimbundu? There was no dictionary structured the way I wanted, searchable the way I wanted, open the way I wanted. So I built kimbundu.org. I share every resource I can find on the language with the world. And because I publish those resources, I owe people accuracy: real people are searching for answers. If PostHog is right, 912 unique users came through last month. ChatGPT is now referring people to my content. I built it for myself, and it turns out a lot of other people needed the same thing.
These projects are not slop. I care about them. What I've stopped caring about is how the code gets made. It took me a while to see that those are different things. People used to argue about programming languages; now they argue about models and harnesses. Same argument, same irrelevance. The thing has to work. Someone has to use it. That's it.
I also want clients
I built PulseEvent because I could build PulseEvent. But I also want people to pay me for it. I want recurring revenue. I want clients. That's why I built a demo for the Taste London festival.
I had a great time at that festival. As I left, I had to ask myself: what would it look like if I lived the life where these people were my clients?
And then I hit the good news / bad news problem. The good news: I built something that held up at a real event in a few weeks with the help of OpenAI and Anthropic. The bad news: I built it in a few weeks.
I don't know what to do with that.
I could always achieve this level of work, but it used to take months. Now it takes weeks. What would someone pay for work delivered at that speed? Is the honest answer a Fiverr storefront, or is there a way to find clients on my own terms? I don't know yet. I haven't landed the first one.
Two paths
One path: chase clients, build a business, deal with the anxiety of sales and pricing and pipelines and "what's the market rate for something that took me a week?" Keep grinding until enough people pay me enough money that I can do this full-time.
The other path: keep my day job, keep building whatever I want, and stop worrying about whether it generates revenue. I paid for these subscriptions anyway. As long as I'm employed, all is well. A handful of projects that bring in a couple grand a month would be amazing. But if that never happens, I'm still building things I want to use, with people I want to build with, on my own terms.
I don't know which path is right. I suspect the answer involves doing both for a while and seeing which one feels less like work.
There's a bigger fear underneath
Two years ago I wrote a post called AI will not take your tech job. I argued the panic was overblown. The hiring slump was economics, not AI. In my defence, the AI of 2024 was a very good autocomplete. What I didn't price in was the loop: agents that pursue a goal, feed their own output back in, and keep going until they hit success or break. Technically still autocomplete, sure. The way GTA 6 is technically a bunch of if statements. So I'm rethinking the whole post. Starting with my own job.
I watch product teams shrink. I can now do what used to take a team, which means a team can be replaced by someone like me, and someone like me can be replaced by whoever asks the models better questions. How employable will I be this time next year? I've joked about learning plumbing if things get bad. But aren't they already bad?
I don't know how to feel secure enough to get a mortgage when the floor keeps dropping out from under the entire industry. I do all of this so I can retire happily in Portugal someday. Some days I wonder why I don't just leave now. Be a farmer. Learn plumbing. Skip the decade of anxiety and get to the part where I have a life.
Until I know
I don't have an answer. I don't think anyone does.
What I do know: code is cheap now. That is not a crisis. It's leverage I didn't have two years ago. I can build a fighting game, a card game, a dictionary, and an event platform simultaneously while employed full-time. When I show up at a festival, I have a working demo. Months became weeks.
The rest I'm still figuring out: clients, revenue, positioning, what to charge, whether to charge at all, whether any of this will matter in three years.
Until I know more, I guess I'll be prompting games and projects that I care about.
One of my next projects might be a shared repo where my friends and I all make chaotic patches to the same game, just to see what comes out of it. No plan, no roadmap, no launch. Just people I like making something together because the tools finally let us. That probably won't make me any money either. I'm doing it anyway.