
Forty-two short pieces on the discipline of not losing.

How leaders turn AI activity into operating leverage. A working manual for the small company in the age of capable models. Not a forecast — a checklist. What an operator actually changes in the calendar, the contract, the headcount, the next quarter.
The first thing a capable model does to a small company is make headcount feel optional. This is the wrong way to read the situation. The right reading is that headcount is now a question.
A question, in the operator's sense, is a thing you ask yourself every quarter rather than every five years. When models were not capable, you hired analysts because that was how analysis got done. There was no question. Now there is a question, and the answer changes by role, by month, by the maturity of the tool you happen to be paying for.
I will give you the framework I use, which is not original but which has the merit of being short. There are three columns. The first is what the model does cheaper than a junior. The second is what the model does worse than a junior, today, but will probably do cheaper than a junior in six to twelve months. The third is what the model will not do, in your time horizon, because the work is not language.
Most of what an operator does sits in column two. This is the column that determines headcount decisions for the next eighteen months. It is also the column that nobody is honest about, because honesty there means saying 'in six months I will let this person go,' which is a sentence with a moral weight that distorts the analysis.
Andreas Anding is the founder and editor of Subversion Press. He runs Remote Native GmbH, a Munich studio building tools for distributed teams. His books — including The AI Operating System and The AI Agent Gap — are made the way every Subversion Press title is: he sets the concept and direction, a language model drafts, and humans edit, fact-check, and stand behind every word.