There is a whole economy now built on teaching founders about AI. Courses, cohorts, certifications, a hundred newsletters promising the twelve prompts that will change your business. Most of it, however well meant, hands a busy owner a second job.
You did not set out to become a prompt engineer. You are running a company. The last thing you need is another skill to learn on evenings you do not have. And yet the pitch keeps landing, because it sounds responsible: get literate, get ahead, do not be left behind.
I think it is mostly backwards. The point of AI in a business was never for you to learn the tool. The point is the work getting done, at a level and a speed you could not reach before, without you doing it. That is a very different promise, and it needs a very different kind of help.
So I do not teach it. I install it. My team goes into the business, finds the work that eats your best people’s time, and builds the system that does it. Not a workshop. Not a framework you have to apply yourself later. The thing itself, running, inside your tools, handed over when it works.
The pattern is almost always the same. Somewhere in your team there is a person who does a task brilliantly: qualifies the lead, drafts the proposal, reconciles the report, spots the problem before it lands. They do it by instinct built over years. The rest of the team does it slower and worse, and the business quietly pays for the gap every single day. That instinct can be captured. Most of it can be built into software that lets the rest of the team perform closer to your best person, at a fraction of the cost and none of the fatigue.
That is the work I care about. Not chatbots for the sake of a press release. Not automation that impresses at a demo and breaks in week two. The unglamorous, specific, high-value work that takes real hours off real people and gives them back to the business.
And I am honest about where it does not belong. Plenty of what gets sold as an AI opportunity is a process problem wearing a costume. If the workflow is broken, automating it just breaks it faster. Sometimes the right answer is not a model at all. A good build tells you that. A course never will, because the course needs you to believe the tool is the answer.
Here is the test I would apply to any AI help you are offered. At the end of it, are you holding a new skill you now have to maintain, or a system that quietly does the work whether you think about it or not. The first is a job. The second is what you actually wanted. I only do the second.