Where I left off

I took a few months off. No meetings, no features, no AI. Just a bag and a lot of moving around.

When I started catching up, my first thought was: well, obviously. The second was: okay, how far behind am I exactly?

Those months away gave me something I didn't know I needed. Perspective with no agenda. No sprint to ship for, no team to align with. Just time to think about where this is all going. And coming back with fresh eyes, a few things hit differently.

The clearest one: build for where the models are going, not where they are today. It sounds obvious. But most of what's being shipped ignores it. There's a flood of AI-wrapped products that are basically a thin layer on top of whatever the current model can do. We build for right now. And right now keeps changing faster than the products do.

Here's the thing that made me stop.

Frontier models can already handle multi-hour software engineering tasks about half the time. That number is climbing fast. If the trajectory holds, we're looking at models running full-day tasks within a year or two.

That changes something fundamental.

The products we design today assume a human is sitting there, guiding in real time. But if a model can work for hours on its own, the whole interaction model shifts. You're not prompting anymore. You're dispatching. You check in, you steer, you course-correct. Status dashboards instead of chat interfaces. That's a very different product to build.

The scaffolding you build around a model has a shorter shelf life than you think, too. Custom retrieval pipelines, elaborate prompt chains, workarounds for limitations that no longer exist by the time you ship. Models improve and the scaffolding becomes dead weight. Build the thinnest layer you can. If most of your product's value is in the workarounds, you're in a fragile place.

And where the real opportunity is, I think, isn't in the demos. It's in the boring stuff. Intake forms. Compliance checks. Support scripts. The repetitive work that runs most companies, most of which isn't in tech and doesn't make for good conference talks. That's where AI is going to land hardest. And most of it is still wide open.

I went away an engineer who builds products. I came back with better questions about what's worth building. That shift in thinking is what the sabbatical actually gave me.

Now I'm ready to put it to use.