11/10/2026

What AI Didn't Do

What can't AI do for a small business?

A line fence with a small gear on one side and a single key on the other

It can't know what only you know, it can't be accountable for a decision, and it can't supply your own experience. Everything else I've handed it, I check. These three stay with me, however good the tools get.

Most writing about AI lists what it can do. I use it every day and I'm enthusiastic about it, so here's the other list: the work it was supposed to change in my businesses that it didn't, and why I'm glad.

What stayed with me

The work What AI does What I keep
Deciding scope, money and names Lays out the options and recommends one The call itself
Writing an article Produces the rough first draft Editing it into my own voice, and the decision to publish
Replying to a customer after first contact Sorts and researches ahead of time The reply itself
Facts from my own experience Nothing. It doesn't have them Every number, story and result
Pictures from my own past Nothing My own photographs
Trusting a safety check Writes the check Breaking it on purpose to see that it works

The common thread is that AI is good at preparing and bad at being the one who answers for the result.

Four things it didn't do, and why

It didn't decide what to publish

An AI-written article is a rough first draft, and a new one never goes live in my businesses until I've read it. My publishing tool is built that way on purpose. It also refuses to publish a draft that still has a marker where only I can supply something, such as a number from my own work or a story that actually happened to me. I didn't want a system that could fill those gaps with plausible inventions.

It didn't replace my own experience

The pictures from my own past are my own photographs, and I kept it that way. I wrote about one of them in Rescuing the Photos I Took From a Helicopter. An AI can generate something that looks convincing, and that is exactly the problem. The real frames are the point.

It didn't make the calls that are mine

I kept scope, money, names and what ships. The AI makes technical judgment calls and explains them, which I covered in Which Decisions to Hand AI, and Which to Keep. If I let it decide those too, I'd be an owner in name only.

It didn't talk to my customers for me

Once someone has become a lead, I write to them myself. I don't send impersonal, templated, cookie-cutter replies. A person who reaches out wants an answer from a person. AI can still sort and research ahead of time, as in How I Help Friends Ease Into Relying on AI, but the reply is mine.

The exception is a plain, all-business message, such as a confirmation from a system or anything that is officially a response from a system. Those work in the other direction: a system's receipt is evidence, which beats "take my word for it." So it cuts both ways. Personal where a person is wanted, automatic where proof is wanted.

Where it was wrong, and what that taught me

AI systems can state wrong things confidently. Anthropic's own guidance on reducing made-up answers lists asking for sources and letting the model say it doesn't know, which only helps if you then check. That's why I keep the checking, and why I try to break the safeguards on purpose, as in Trust, but Break It on Purpose.

One lesson I keep coming back to: asking a free or lower-quality AI to do a job above its pay grade can cause more problems, glitches and bugs than the job was worth. If it takes longer to untangle than it would have taken by hand, doing it by hand was the better deal. Match the tool to the task.

Review whatever an AI builds as thoroughly as your goals and time allow. That review is yours too.

What it did change

I don't want this to read as a complaint. AI changed a lot: the first draft of nearly everything, the routine checks, the chores nobody wanted. Where it fits is in preparation, and I wrote about how to start in How I Help Friends Ease Into Relying on AI.

The point of this piece is to be accurate about the boundary. The people who get the most from AI are the ones who know exactly what they're keeping.

FAQ

What should a business owner never hand to AI?

Decisions that depend on what only you know, anything that commits money, and anything a customer will see before you've read it.

Will AI replace the owner's judgment?

It can prepare and recommend. The accountability stays with you.

Does AI make things up?

It can state wrong things confidently, so check anything that matters, especially numbers and sources.

Should AI reply to my customers?

Not once they've become a lead. Write those replies yourself. Automatic messages are fine for confirmations and receipts, where a system's record is the point.

Is a free AI tool good enough?

For simple, low-risk jobs, often yes. For work above its pay grade, it can cost you more in glitches and fixes than doing the job by hand.

Let my Claude help your Claude

Working out which parts of your business stay with you is exactly what my one-on-one AI guidance sessions are for.

Book a free 30-minute consult

Researched and drafted with AI, then reviewed and edited by the site's editors.