AI makes it easy to say yes to complexity.
We can build that feature. We can add another agent. We can create tooling to orchestrate the agents managing the tooling. I’m excited about what has become possible, but I still believe the simple solution wins in the end.
I once had to maintain 22 production applications by myself. I designed and built an elaborate workflow for coordinating maintenance across all those repositories. It did what I had designed it to do, but every time I had to use it something broke. Of course AI can help me fix it, but then I’m still stuck fixing the tooling and not the product.
The better solution turned out to be a simple skill and subagents. The skill described the work, the agents applied it across the repositories, and there was much less machinery to maintain. It was simpler, more flexible and easier to change when something stopped making sense.
That may be why terminals, CLIs and skills keep surviving every new wave of AI tooling. Because they are simple, they are easy to replace or adapt when you switch providers. And almost always they are all you need.