Mechanic or driver? You don't need to understand AI, but you do need to read the dashboard

Most business owners I meet are trying to become AI mechanics: learning prompts at midnight, comparing models they'll never need, watching tutorials instead of running their business.

There's a better way to think about it. Are you the mechanic, or the driver?

A mechanic knows how the engine works and, when something breaks, knows where to look. A driver gets from A to B without opening the bonnet. With AI the same split exists. A few people build it. Most people just use it.

Both are fine. You don't need to be an engineer to run a business. But knowing which one you are changes what you ask, who you hire, and how you judge the tools in front of you.

The E-Myth, updated for AI

In The E-Myth Revisited, Michael Gerber describes the three people inside every small business owner: the technician who does the work, the manager who organises it, and the entrepreneur who plans the future. His famous advice is to work on your business, not just in it.

AI adds a fourth job to Gerber's three: someone has to look after the AI itself. Many owners take that job on as the mechanic: learning prompts, comparing models, watching tutorials at midnight. That's Gerber's technician trap again, working in the business instead of on it, just with new tools.

My twist is simple. Be the driver, and insist on a proper dashboard. A driver doesn't need to understand fuel injection. They need a fuel gauge, a speedometer and a warning light, and they need to act when one of them changes.

I'm a mechanic, and I still rely on gauges

I'm a machine learning engineer, and I spent years building backend systems that handled peaks of a quarter of a million requests a minute. I can open the bonnet. But in my own businesses I rarely do, because I build dashboards that tell me when I need to.

• In The Modern Shrine, every AI-written report is logged to a private dashboard showing the success rate, failures, response time and cost per report. In the 30 days to 22 August 2026, it showed 75 of 82 free report previews passing the quality checks, and 7 stopped before a customer saw them. I look at it; I don't dig through logs.
• In the MummyReady funnel, every step fires a tracking event, so the team could see where visitors dropped out at each question.
• For my ad campaigns, I kept a log that pinned every result to the exact version of the website that was live, so I'd never mistake a website change for a keyword success. The Google Ads story.

If a mechanic needs gauges, a driver definitely does.

The four gauges

These are the four I aim to put on every AI system I build for a client. If your vendor can't show you these, you're driving without a dashboard.

1. Cost per task

   Not the monthly bill: the cost of each enquiry answered, each report written, each lead scored. A figure like, for example, four cents per enquiry is a number you can make decisions with. A surprise invoice is not.

2. Error rate

   How often the output is wrong, sent back or corrected. It should be tracked from real work, not claimed from a demo.

3. Handovers

   How often the system says "I'm not sure" and passes to a person. Too high and it isn't saving time. Zero is suspicious, because it probably means the system is guessing instead of asking.

4. Owner

   One named person responsible for checking the gauges each week. A dashboard nobody looks at is decoration.

What drivers should ask, and what they can ignore

Worth asking:

• What does each task cost, and what happens to that cost if volume doubles?
• How do we know when it's wrong?
• What happens when it isn't sure?
• Who do we call when a gauge turns red?

Safe to ignore:

• Which model it uses, unless cost or privacy depends on it.
• Parameter counts, benchmark scores and "agentic" architecture diagrams.
• Most AI news headlines. (I'll translate the important ones in AI headlines, translated.)

Where a mechanic still helps

Drivers still need a good mechanic sometimes: when setting the system up, when a gauge turns red, and when deciding whether a new tool is worth switching to. That's the role I play for clients. I build the engine, fit the dashboard and teach the owner to read it, so they can stay in the driver's seat.