AI in 2026

I’ve been using a significant amount of AI generation in my coding in 2026. I used bits and pieces before, but not at this scale. I’m not using agents, because I can’t afford the fallout from their more catastrophic failures, like deleting codebases or databases. That’s just not something I want in my life.

There’s also a data integrity aspect. Since I’m working with very sensitive data, I don’t necessarily want that data available to or uploaded to external servers. I have my own tools for providing subsets of files and things like that, which lets me control the entire data flow.

The other benefit is cost. I get working code for significant functionality out of a few thousand tokens, if that. That’s insane to me compared to people using millions or billions of tokens to create incomprehensible nonsense.

One of the major things this has highlighted, now that I’m working predominantly on my own software, is why other people who’ve been in this position for the past two or three years haven’t shipped much at all. I’ve released five new products so far this year. All of them have been internal or augmentations to my main service, but that’s predominantly because I’ve been stuck dealing with a lot of things that reduced my productivity anyway. It’s only been during March that I’ve been able to be fully active on this work and my productivity has skyrocketed.

I’ll be operating on my main product almost full-time now and I’ll see how far I can push it now that I have the full AI toolset. So I’m finding a lot of value in it, but I’m not working the same way everyone else is posting about. A lot of guys are using agents. I’m not. They say they’re shipping a lot of things, but I’m not seeing a massive amount of it.

So it makes me think about why there’s this difference. My view is simply that AI is not going to solve-and is likely going to exacerbate-the issues that come with having multiple people working on things. For individual productivity, it can be brilliant, as I’m finding. But for team-based work, you still get the endless smoothing-over that comes when you add more people, average out their individual productivity and absorb all that friction. AI does not help with that. If anything, it probably makes it worse.

And I think that’s why large, arguably overstuffed organisations like Microsoft and GitHub are seeing their effectiveness and product stability plummet to laughable levels, while quiet individuals are just getting things done.

Having come predominantly from the automation space, I settled into AI quite well anyway, because it’s effectively just pattern recognition leading to the automation of text generation. Coding is effectively text generation, so it’s good at it. Some logical chains are in there as well, because text prediction can also capture the logical chain embedded in text.

But that’s as far as it goes. It will not help you move strategically, because strategy isn’t really written down that much. It’s experience. It’s tacit knowledge that operates in experienced people’s brains. Real strategy usually isn’t written down. It’s defended and it’s situational.

Additionally, the things we find painful-the things that contribute to our overall personal strategic approach-are not consistent enough as a broad pattern across a language for them to be cleanly represented. In less complex terms, the things I hate shape my strategy and the things I don’t hate as much generally don’t.

For example, I hate unnecessary, needless work. I absolutely despise it. It’s one of the things my personality reacts to heavily. So I won’t do anything in Java, because the idea of Maven or other build tools refusing to compile standard code for whatever reason feels like unnecessary work to me. It feels like I’m babysitting my tools rather than my tools just doing their damn job. Same deal with Java virtual machine management and a bunch of other considerations as well.

Hence, the majority of my web services are written in static HTML with other tools and languages that are more efficient to work in than Java. Same deal with package management. Other people love using community packages because it lets them write code more efficiently. But I absolutely hate the maintenance aspect and I’m experienced enough to understand the threat aspect that comes with that. So I stick to core libraries or write my own libraries to standardise what I need and let me stay selective.

Those standard libraries and core libraries mean that, in all my time, I’ve never had a supply chain issue, because I’m that selective about what I use. That is my strategy, or at least a small aspect of it and it’s highly personal built on personal distaste-even open hate-as well as many stressful, horrid experiences. My pattern recognition is pretty good after all that, probably my best tool.