AI is a tool, not a strategy

Think of a kitchen

Every station in a kitchen has its utensils, and each one has a job. The rice cooker makes the same rice every time you press the button. The knife does the prep. The chef can do every one of those jobs by hand, but slowly, and the chef is the most expensive person in the building.

The utensils only change how fast the food gets out and how much it costs to get it out. The dinner still has to reach the table either way.

The tools change the speed, not the job

AI is one of the utensils. Many of the jobs people hand to AI are better done by a plain bash or Python script, or a workflow tool, because the steps never change. Use AI where a step needs judgment, and only once you know what the job is.

The people getting the most value out of AI spend less time studying what AI can do and more time understanding the work that has to be delivered: what has to happen, in what order, and what has to be in place for it to get done. Then they work backwards to the tool.

Strategy is still your job

Say you have a lead problem. AI will not tell you which people to go after, where they are, or how to keep them warm once they find you. It will give you a long list of ideas. You still have to architect the pipeline yourself: where leads come from, what happens when one arrives, and what a good one looks like. Once that is designed, tools like a CRM automation, and AI where it helps, can run it. If you skip the design, you spend your time working behind Claude, hoping the thing it built works.

Before you automate anything

1. Write down the work that has to get delivered.
2. Write down every step, and who or what does it today.
3. Mark the steps that need judgment. Those are the only candidates for AI.
4. Script the rest.

That split has a name. It is the difference between a deterministic, an autonomous and an agentic workflow, and the three AI workflow types explains each one with an example.