I don't think the world needs another article about AI and programming, but I want to capture this moment in history, mostly for myself.

First of all, I consider myself incredibly lucky to have lived through a historical period that allowed me to experience the birth of three complete revolutions: the internet, smartphones, and now LLMs.

The acceleration I've experienced toward using AI has been nothing short of staggering. After more than a decade as a programmer, during which nothing fundamentally changed in my workflow, everything has been completely upended in just a couple of years — and particularly in these last few months. The simple truth is that, despite still doing the same job, I barely write code anymore. And that's something I never expected to witness.

ROI

In a way, it's an exhilarating, liberating feeling. You can move so much faster, test ideas in a tenth of the time. There are entire experiments I've completed recently that I wouldn't have even dreamed of starting. The effort required to set up an environment, climb the initial learning curve, iterate… it was enough to kill any mental ROI calculation before I even began. Now those same projects take just a few hours.

Here are some concrete examples of curiosities I've been able to explore, only thanks to LLMs:

  • Over the Christmas holidays, I released two native Ruby gems (mquickjs-ruby and quickjsruby) that provide a complete JS runtime. From idea to finished project with tests, benchmarks, and documentation — in a couple of days. With about twenty prompts. From my phone, without a laptop.

  • I was looking for an app to share a shopping list with Taiza. Simple and synced. I couldn't find one that was dumb and basic enough. In a few hours, I released Noto (GitHub). Anonymous, dark/light mode, super clean interface. Among other things, it uses Supabase — a product I'd never tried before.

  • I converted this blog from SvelteKit to Astro, something I'd been wanting to do for at least a year, even though the site had no real problems. One of those classic tasks that's tedious and hard to justify… unless an LLM does it for you. In a few hours.

  • At work, we're working on a new feature launching soon. One morning I thought, "Wouldn't it be great to have a motion design presentation video?" I discovered Remotion, a tool for creating videos programmatically. I'd never heard of it. The bootstrapped project already has skills ready to go. I iterated with Claude on the storyboard and then had it write all the necessary code. By that same morning, the video was done.

All of this is absolutely wild, and I still can't fully wrap my head around it.

I don't even feel like the quality of the code I produce is declining: I read the generated code carefully, verify everything, and spend considerable time specifically asking for refactors and renames until the result matches my taste. I remain in total control, just like before — but with timelines that are a fraction of what I could achieve alone. And I consider myself an extremely fast programmer.

Manager of agents

The biggest annoyance I've noticed is that the flow is much more fragmented. When programming, I used to be "in the zone." Now I ask for something, then wait.

Maybe not for long, but long enough to break the flow — and for other ideas to start floating in, distracting me. What I often end up doing is trying to parallelize two tasks, the current one and the new one, with two agents. And that's the beginning of the end: I transform from a developer "in pair mode" into a manager of agents, with all the cognitive stress that brings. I don't like managing people. I like to program.

Open questions

It's all extremely new, and I still need to adapt. The unanswered questions are endless.

  • If code is basically free now, will SaaS as a concept still be a thing in the next decade? What kind of products will survive?

  • Will high-level code just be the next Assembly? Will we really lose complete control over code and the processes to write it and just focus on the end result, like Gas Town and similar AI-orchestration projects are trying to prove?

  • More broadly speaking, will humans still need to work? Which human skills still matter?