Nobody really knows anything. Ship and be happy.

Every week there is a new correct way to work with agents. Prompt engineering, then MCPs, then skills, sub-agents, workflows, loops, and whatever gets named tomorrow. The vocabulary sounds invented because most of it is. It is also, somehow, all real.

Underneath the tooling is an identity problem. Engineers used to introduce themselves by what they built, and that sentence carried weight. It carries less now that anyone can say “I built an app myself.” So the differentiator moves somewhere else: I burn more tokens than you. My stack is more complicated than yours. Therefore I ship more.

There is an industry selling into that insecurity. Not snake oil anymore—just someone’s personal flavor of process, packaged as the thing you are missing.

Early in the cycle, companies sold fine-tuned models that would beat the frontier at one narrow task. You did not have to beat them. You only had to wait. The next general model arrived and did the job well enough, and the one after that did it better. The same is true of the scaffolding built around the models.

So most of the time you are better off typing the prompt yourself into the best model available than spending the week learning this month’s workflow.

This is not an argument against building tools. It is an argument against buying someone else’s. Prompt the model, find the gap in your own work, and build only that. Software has never been more personal.

If you feel behind, you are not. The game changes daily, kingdoms rise and fall, and there is no secret workflow standing between you and the work.

Pick a model. Build the thing that brings you joy. Let the timeline sort itself out.

Original: burcs.dev