Working with it

The conversation around AI in software development is loud and usually binary. Either it replaces everything and we’re all out of jobs, or it’s overhyped and barely useful. Neither take ages well.

I’ve been thinking about how to approach this without just reacting to it. Not ignoring it, not panicking about it — figuring out what it actually means for the way I work.

The first thing I noticed: fear is a bad compass. There’s a version of this where you drop what you know, chase whatever feels safe, and end up somewhere that doesn’t fit. A stack you don’t care about. A direction that isn’t yours. That might look like adaptation. It’s usually just anxiety wearing a plan.

So I went the other direction. I leaned into what was already mine: architecture, system design, cloud. Not because it seemed safe, but because it’s where I was headed anyway. And it turns out — the parts of software development that AI handles least well are the parts that require judgment. What to build, how to structure it, where the system breaks under load, which tradeoff is worth making in this context. Those aren’t going anywhere.

The most important thing I’ve internalized working on recent projects: no dogma. Every architectural decision is individual. What works at one scale, in one team, with one set of constraints, will be wrong somewhere else. That applies to software design and it applies to how you use AI — there’s no universal workflow, only context.

In practice, AI works well when I’ve done the thinking first. Scope defined, requirements clear, edge cases identified. When the input is precise, the output is close. Small review, small adjustment. When the input is vague, the output is too — same as any other tool.

It’s also useful as a sparring partner when there’s no one else in the room. Talking through an architectural decision, stress-testing an assumption. It’s not a replacement for a senior engineer or a team with real experience, but it’s better than silence. The grain of salt is mandatory.

Documentation surprised me. It doesn’t get skipped anymore. The cost of writing it dropped enough that it actually happens.

I’m not drawing a final conclusion here. The situation is still moving. But the approach seems right: stay curious, stay grounded, don’t let the noise make decisions for you.