# AI Isn't as Smart as My People. So Why Use It?

Source: https://metabinary.ie/blog/ai-not-as-smart-as-my-people

[← Back to blog](/blog)

AI Reality Check

# AI Isn't as Smart as My People. So Why Use It?

By [Joseph Rose](/about), founder of MetaBinary. 20+ years building enterprise software. Built the world's first instant online metal-casting costing engine.

Published 2026-10-22

Your best estimator knows things about your jobs that no AI ever will. Keep it that way. The point of AI is to clear their desk: the lookups, the re-typing and the report-building that eat their week. Then their hours go to the calls only they can make. When I built a costing engine from estimators’ know-how, it made them faster, and they stayed the ones making the calls. Your experts’ time is the scarce thing, and AI is how you buy more of it.

## You are right about your people

An estimator is the person who prices a job before you send the quote. A good one looks at a drawing and knows which feature will cause trouble, which customer always changes the spec, and which machine to keep free. Years on the job taught them that. None of it is on the internet, so no AI model has learned it.

So when an owner tells me their people are smarter than any AI, I agree. On the things that matter most to your business, they are. AI doesn’t know your customers, your machines, or your margins.

Judging the model on its own misses the work around it, as I explain in [why an AI project costs real money](/blog/why-ai-projects-cost-money). The more useful thing to look at is how much of your best person’s week actually needs their judgment.

## Most of an expert’s week is groundwork

Watch an expert at work and a lot of the day is lookups. They search for the last time you made a similar part. They copy figures from one system into another. They chase someone for a number. They build the same report every Monday.

I talk to a lot of manufacturers at expos, and two things come up again and again. The same data gets typed into two or three systems. And the know-how for quoting sits with one or two people.

None of that groundwork needs your estimator’s years of experience. It just needs doing, and right now your most skilled person is the one doing it.

## AI is good at finding, sorting and drafting

This is the work AI does well. It can search years of old jobs in seconds and pull out the ones that match. It can read a supplier’s email and put the numbers in the right boxes. It can draft the Monday report, so your expert only has to check it.

It is poor at the calls your expert makes. New situations, unusual customers, and trade-offs with no clear rule trip it up.

[A study by Harvard Business School and Boston Consulting Group](https://www.ssrn.com/abstract=4573321) showed both sides. In it, 758 consultants did realistic work tasks, some with GPT-4 and some without. On tasks the AI could handle, the consultants using it finished 12.2% more tasks, worked 25.1% faster, and turned in better work. Strong performers gained as well as weaker ones. On one task picked to be beyond the AI, the consultants using it were 19 percentage points less likely to get the right answer.

The same tool made skilled people better at one job and worse at another that looked just as hard. Sometimes AI gets it right on the first try, and sometimes it never does, however you ask. [Why AI gets it wrong the first time](/blog/why-ai-gets-it-wrong-first-time) covers how to tell which kind of task you have.

## Your expert stays in charge

The way to use AI safely follows from that study. The AI does the groundwork and shows where every answer came from. Your expert reads it, checks anything that looks off, and makes the call. If something needs to be sent, bought, or changed, a person says yes first. That is how we build on the [AI Builds page](/ai): answers show their source, and a person stays in charge.

My own costing engine worked the same way. I built the world’s first instant online metal-casting costing engine. The hard part was getting estimators’ know-how into rules a computer could follow. Once that was done, a price came back instantly, and the same part got the same price every time. The estimators stayed. They got faster, and they still made the calls on the odd jobs and the key customers.

A carefully built tool in the middle helps a lot. The AI asks and explains, and the tool works out the real figure. [Why good tools beat a smarter AI](/blog/good-tools-beat-smarter-ai) explains how that works. And when something does go wrong, a person answers for it, which is the subject of [who is responsible when AI gets it wrong](/blog/who-is-responsible-when-ai-gets-it-wrong).

## The bigger risk is know-how nobody wrote down

If your best estimator’s rules live only in their head, AI can’t use them. Nobody else can either. When that person is off sick, on holiday, or retired, quotes slow down or go out wrong.

So the first step is often writing that know-how down: the checks they always make, the customers they treat differently, the jobs they will not quote below a certain size. Once it is on paper, a new hire can learn it and an AI can follow it. [Process Capture](/automation) is how we do that.

AI also isn’t for everyone. If your experts already spend most of their day on judgment calls, and the lookups and re-typing are light, AI will not buy you much time. Put your money somewhere else.

## Start with your expert’s slowest task

Pick one task that eats hours of your best person’s week and doesn’t need their judgment. Finding old jobs, pulling numbers together, or building a regular report are good places to look. Build a small version on your real data and let your expert try it. If it frees up their time, build it properly. If it doesn’t, stop there and keep what you learned.

**Send me your expert’s slowest task.** Tell me what takes your best person the most hours through the [AI Builds page](/ai), and I’ll tell you if AI can take it off their desk.

## Common questions

### Will AI replace our staff?

Not the people who make the judgment calls. AI is good at the groundwork around their work: finding old jobs, pulling numbers, and drafting reports. When I built a costing engine from estimators' know-how, it made them faster, and they still made the calls.

### What is AI actually good at?

Finding, sorting, and drafting. It can search years of records in seconds, pull out the ones that match, and write a first draft for a person to check. It is weak at new situations and at calls that need to know your customers and your shop.

### How do we know when to trust it?

Test it on questions you already know the answers to, and count how often it is right. Use it where every answer shows its source, so your expert can check it fast. Anything that sends, buys, or changes a record should wait for a person to say yes.

## Where to go next

-   [If AI Can Do Anything, Why Does an AI Project Cost Real Money?Running an AI model costs little. The money goes into picking the right question, connecting your data, building the tools the AI uses, and checking its answers.](/blog/why-ai-projects-cost-money)
-   [Why Didn't AI Get It Right the First Time?AI gets general jobs right first time. Jobs that need your own rules fail until someone writes those rules down and tests the AI against real answers.](/blog/why-ai-gets-it-wrong-first-time)
-   [Who Is Responsible When AI Gets It Wrong?Courts say the business that uses the AI is responsible. So keep a person signing off on anything your AI sends, buys or changes.](/blog/who-is-responsible-when-ai-gets-it-wrong)
-   [AI BuildsWhat we build, and how a build runs.](/ai)