This week’s posts converge on a single claim: AI only works in your favor when you control it.

Local auto-completion shows this at the smallest scale. When the model runs nearby, response time drops, data stays put, and behavior is predictable. You see cause and effect directly. That changes how you write and think.

Running your own LLM makes the same principle explicit. Control removes mystique. The model stops being an authority and becomes software. You can inspect it, test it, break it, and decide where it fails.

Comparing models reinforces that control requires choice. There is no universally “best” LLM. Different models expose different limits. Without comparison, selection turns into belief. With comparison, it becomes engineering.

Moving beyond your own hardware does not contradict this. Cloud and hybrid setups can still preserve control if constraints are explicit and reversible. The issue is not location. The issue is whether you can leave.

The dividing line is simple. If you can inspect it, compare it, and move it, AI is a tool. If you cannot, it is a dependency.

Have a great weekend, everyone!

Join my AI Pioneers Club today to get premium features such as AI workflows, my AI toolbox and 1:1 calls. Or reply to this email to arrange for personal tutoring or if you want to discuss a project with me.