Somewhere there is a slide titled AI strategy. It lists a summarizer, a copilot, chat with the documents, chat with the data, a bot on the website, and a promise to revisit agents next quarter. Each item has an owner in the sense that a name sits beside it. None of them has an owner in the sense that a person would notice if it disappeared.
This is not ambition. It is a refusal to choose.
one job, learned
Pick the workflow that actually hurts. Put one tool in it. Stay until the people who do that job are faster or clearer, and until they can tell you, without a script, what the tool is bad at. That last part matters. A team that can describe the limits will use the tool. A team that has been told it does everything will try it once.
Slower here means sequenced, not stalled. You can move quickly inside one job. You cannot move quickly across nine and learn any of them.
Your users will trust a tool you were honest about. They will not trust a pile.
subtraction is a strategy
Turn off the experiments nobody returned to. An unused bot on a site teaches customers that you do not mind wasting their time. An unused copilot teaches employees the same lesson about you. Removing it is a product decision, and it makes the thing you keep easier to see.
The companies that get value are not the ones who announced the most. They are the ones who can point at a Tuesday that is different, and can say why they have not shipped the other eight.