Before and After AI

We'll remember this period as a turning point, dividing the time before AI from what came after.

It's hard to say whether this revolution will be an opportunity or a threat, a good thing or a bad thing, for our society and for humanity as a whole. I think it's still hard to make a prediction, especially given how quickly AI is evolving.

For my part, I have doubts, as I think most of us do, but also high hopes. The nerd in me loves it almost unconditionally. I'm fascinated by everything it lets me do, both at work and in my personal life.

Programming in an AI World

Being able to rewrite an entire modern web interface, hardly the most enjoyable task, in about three hours (more than 20,000 lines of code) is incredible. It would have been unthinkable just over a year ago, partly because the early LLMs were nowhere near today's models when it came to code quality, following instructions, and so on. If you put together a good plan, you're already halfway there—and then some.

I don't want what I've just written to reinforce the idea that programmers are no longer needed. What I would say is that programmers of the future won't write code the way they used to. They'll need to focus more on building their skills and putting them to work alongside the ideas needed to achieve a particular goal. That's no small thing, of course. Skills can be learned in many ways. Ideas, on the other hand, depend partly on knowledge of the field you're working in and partly, I think, on something deeply personal. Let's say that the ideas that really make a difference come from relatively few people. At the same time, I believe passion and determination can play a fundamental role too.

My experience with LLMs over the past year has mattered a great deal to me. They've helped me complete many projects at work. I also encouraged colleagues to use them, bringing a breath of fresh air and pushing the company to adopt them, rather than the other way around.

Here's a silly example: think about how much time and energy you no longer have to "waste" on all sorts of configuration (Docker, CI workflows, shell scripts, and so on). You can focus on the actual project instead. That's priceless. I don't know about you, but being able to skip that pain-in-the-ass work has been a godsend for me.

The Appeal of Local Models

Right now, my only regret is that I haven't managed to try running models locally because the hardware costs an absurd amount—it's practically priced like gold.

So, while OpenAI's and Anthropic's frontier models are currently the best and readily accessible, Chinese models released for free to run locally have also made great strides. You can see it in the many open source projects built around them, including the now-famous DwarfStar.

Being able to break free from subscriptions, usage limits, and private companies, while gaining more privacy and no longer feeling tied to a service, would certainly be a good thing. It's a goal worth pursuing.

A Team at Our Disposal

Another thing I've been thinking about lately is how we use AI. There's a risk we'll focus on what it can do for us and forget that, with our own input and knowledge, it could help us overcome some of the physical and practical limits on what we can achieve. It's become a great partner to work with—almost like having a whole team at our disposal.