Some Thoughts on AI, While Things Are Still Unsettled

To start with, I don’t pretend to have answers.

I also tend to think that predictions mostly say something about the person making them, not about what will actually happen.

I’m just putting a snapshot somewhere. Something I might come back to later, to see what I was thinking while all of this was still unclear.

The feeling I miss from software development

For a long time, the best part of software development was the act of making something.

You would put effort into solving a problem, struggle with it, design something, implement it, and then see it run. You could use it, break it, fix it. There was a direct link between effort and result. That link mattered.

Seeing something operate because you made it operate was satisfying in a very concrete way.

With AI, we’re still getting that satisfaction, at least for now. Prompting, watching code being generated, seeing things work almost instantly. But it comes with a side glance at how necessary we actually are in the process, and where this is going.

It feels like software development has been TikTokified. Fast gratification. Less friction. Less depth. It can feel exciting, but something about it feels wrong. I don’t know how long this phase lasts. I just know it doesn’t feel solid.

Building on something that won’t stay still

Right now, almost everything is being built on top of AI models.

Tools, workflows, products, integrations. All of it sits on systems that are changing very quickly. You build something today, and tomorrow the model behaves differently. New capabilities appear. Old limitations disappear.

I don’t think this makes current work useless. There is value in building, experimenting, and learning. But it feels temporary. Like building on ground that doesn’t stay in place long enough to settle.

A lot of what’s being built feels like wrappers around capabilities that haven’t stabilized yet.

Production is accelerating faster than demand

What keeps bothering me is the imbalance between production and demand.

AI lets us produce far more than before, we’ve got more code, more content, more features, more products. But demand doesn’t move at the same pace.

Historically, productivity gains haven’t led to people working less or transitioning smoothly into other kinds of work. Companies optimize for cost. They reduce headcount. They do more with fewer people.

I don’t see a clear reason why this would be different this time.

Productivity doesn’t automatically create meaning. It doesn’t automatically create need. And it doesn’t guarantee a soft transition for the people displaced by it.

Replacement feels irreversible

I do believe that if I can do something well enough, cheaper, and faster, it will eventually do it. Once that happens, the old version doesn’t come back. This seems especially true in software.

I started using AI early, and the combination of AI tools and developer expertise was incredibly powerful. I’ve watched the models and surrounding tools evolve. What stands out is that the goal switched very early from helping people write make things faster. The direction seems broader: automate every task that can be automated.

I don’t think all this is necessarily bad. Things not being done by humans isn’t inherently a problem even if it feels very uncomfortable and can be source of anxiety.

The real uncomfortable question is what are we doing instead?

And more importantly, who is actually waiting for people to figure that out.

Historically, transitions don’t pause. They happen, and people adapt afterward, or don’t.

What feels heavier, slower, more real

If anything here feels like it might last, it’s not what’s built with AI, but what AI depends on.

Most current products feel like application-layer work, wrappers around rapidly advancing models. They can create short-term value, but they don’t feel durable. As models improve, many of these things become trivial to reproduce or get absorbed by larger platforms with more data, compute, and distribution.

Once capabilities stabilize at a high enough level, it becomes very hard to compete at that layer.

What feels heavier is the infrastructure and everything that allows AI to exist without breaking things.

These things move slower than capabilities. They are harder to replace. They don’t disappear just because models get better.

Even that part of the industry doesn’t have a place for everyone, but it feels more grounded than everything built on top of constantly shifting capabilities.

Right now, that’s the only part that feels somewhat stable to me.

Sitting in the middle of it

I don’t know what direction it will take, everything is still blurry. We’re clearly in the middle of it, not at the beginning and not anywhere close to the end.