The A.I. Mandate

Bastards! They got to me. The AI guys, I mean. So, I was mandated to start using AI at work. At this point, it was still sort of a soft push, but with the current situation on the labour market and my age and the fact I would have to know AI to land anything, anyway, I decided to cave in and try that. Survival mode. But then, a strange thing happened: it kind of worked nicely. I'm not gonna become an AI evangelist now; in fact, I still think it's not sustainable, but "it works on my machine."

My team member went through all the conceptual work and set the workflows, basing them on Matt Pocock's ideas; and it turns out that AI, having access to the codebase and descriptions, is able to produce above-the-average results. I've worked with enough mid-skilled devs in my career to see that it's "good enough." But done faster. When compartmentalised smartly, the programming goals are achievable. I think we have a couple of months like that, at the very best, before the actual pricing kicks in, so let's make use of it while we can, hoping that no one is let go in the meantime, and we can go back to the old ways; we're already in the so-called bootstrapped mode on the tech side, so this would be harmful even with AI.

I can't really speak for non-programming use-cases, because that seems another aspect. I hear people saying they're using AI for research, and that, I wouldn't trust. But low-level, non-business development seems to be working out okay. If anything, it might further standardise the way React (including Next.js) is developed, because the AI is not coming with anything new, just reads from the past.

A friend of mine said that after the year 2000, everything seems magic to non-technical people anyway, so to them, that aspect didn't change much. I also realised that I don't like the AI for the same reason I didn't like all kinds of frameworks before I learned to live with them: it takes away control on large areas of the codebase. It's yet another layer of abstraction on the pile of NPM dependencies that do God knows what. But we're not vibe-coding the entire apps yet, just pieces.

Being a tech lead since 2017, I've spent most of that time working with people either by pair-programming or reviewing their code; which, if you think about it, is asynchronous pair programming. It looks now like a perfect training ground for "developing with AI." And then, reading and writing is another aspect of my activities that prepared me for this situation. I don't feel comfortable with admitting that, but -- again -- "it works on my machine." And it's not, like, my pet project. It's work.

So, we'll see how it goes. I decided that if I have to do that, let's at least have fun while doing it. No sense in getting all worked up about it: it would cost me my nerves, and in the end, I would change nothing. Apart, maybe, from a job, but, as mentioned earlier, this could be tricky in our current landscape. My role is much wider than programming alone, so it doesn't take "all" from me.

But I have to say that it felt as if something died, some sort of art of programming, which will now be delegated to a black box producing plausibly-sounding sentences. As I said, not my personal codebase; I didn't even started it, merely inherited when joining.

On the bright side, I finally get people to demand proper descriptions of what they want, otherwise "computer says no." :D

And because I am using it, I configured it to sound more like I'd want it, which means no sycophancy, no emotional, and using vocabulary back to 1850. Yes, I am still under the spell of "Moby Dick." This led to a funny situation where a colleague and me were developing a feature, and Claude Code kept using phrases like "the smoking gun" or "solution hinges on," and my colleague, having somewhat less knowledge in English, was a bit lost.

This brings another thing, too. So, I am fully aware that it's merely a large language model, not a sentient nor even reasoning being, and I can be impressed with this tech: it catches the tone, etc., really well. It throws in remarks matching the sophistication level of the prompter. And I am not surprised that people fall for that. I mean, hell, even Richard Dawkins decided recently that AI passed the Turing test; not in his own words, of course. As such, we're doomed.

(I should have check how many followers I had, and how much I'm gonna lose now. :D People on Mastodon can be pretty radical about AI, which sentiment I understand and share, even if I'm "working with the other side.")