Your 30-day AI evaluation plan that does not require an engineer

Most leaders I talk to are stuck in the same loop. They know they need to do something about AI. They read the articles, forward posts to their team, attend the conference. Nothing changes, because they are trying to find the right answer before running any experiment. AI does not reward the best reader. It rewards the person willing to run a small test and be wrong in week one.

Here is the plan I actually followed in my restaurant, spread over a month, no engineer, no consultant, no budget approval. It requires a few hours a week and a willingness to make one decision at the end.

Week 1: Find the pain

Do not start with the technology. Start with the thing that hurts every single week.

Make a short list of the recurring work that eats your time or your managers’ time: the report someone assembles by hand, the follow-up emails that slip when it gets busy, the checking and re-checking of yesterday’s numbers. Pick the one that repeats most often. Not the most exciting one. The most boring one. Boring and repeatable is exactly what AI is for.

Then describe it out loud to an AI assistant like you would describe it to a new hire: what goes in, what comes out, what good looks like, where it usually goes wrong. If you can describe it clearly, it is automatable. If you cannot, you found a different problem worth knowing about.

Week 2: Build the draft version

Have the AI actually do the task this week, in draft mode, alongside the human process. Not instead of it. Alongside it.

This is the rule my whole operation runs on: every system drafts, a human decides. Week 2 exists to compare the machine’s draft against what your team produces by hand. Some days the draft will be embarrassing. Refine the instructions the same way you would coach a new team member, and write the refinements down. That document becomes the system.

Week 3: Add the accountability

Before you trust it, give it what any new employee gets: one named owner, a clear definition of done, and a way to shut it off in ten seconds. Decide what it is never allowed to do without a human. Mine never send anything to a customer or a team member on their own until they have earned it, and even then a human holds a veto.

Then measure the only number that matters at this stage: hours. How many hours a week does the draft version give back to you or your managers? Write the number down. You will need it in week 4.

Week 4: Decide what the hours are for

This is the step every 30-day AI plan skips, and it is the whole point.

You now have hours coming back. The default is that busywork expands to eat them and nothing in your life or your team’s life actually changes. The alternative is deciding, on purpose, where those hours go.

My answer is people. The hours my systems return go to the parts of leadership no machine can do: the conversation with the team member who has gone quiet, the thank-you note in someone’s first language, the dinner where I am actually present instead of mentally running the catering order. I even point the AI at that too, with questions like: who on my team deserves recognition that has not received it? The machine finds the name. I do the human part.

That is the full plan. One month, one system, one decision. Automate the ordinary. Then go amplify the extraordinary.

The people questions are numbered and free at jonharmeling.com/prompts. Start your week 4 there.