Can an AI really tell you whether your job is automatable? What Joe can and cannot do

October 1, 2026 · 4 min read

People are right to be suspicious of a chatbot that recommends robots. Here is exactly what the concierge on this site knows, what it is forbidden to say, and where it hands you to a human, written for the reader who has seen 2001 and would like to keep the pod bay doors operable.

The short answer

Yes, within limits that are written down and enforced. The concierge can take a plain description of a station and tell you whether the physics work, which class of machine applies, what the manufacturer-published figures say about specific candidates, what usually disqualifies a shop like yours, and what the planning range looks like. It cannot tell you a cell is safe, cannot quote a fixed price, cannot see your floor, and cannot cite a study it merely remembers. Those are not modesty. They are rules in its instructions, and they exist because a confident wrong answer would destroy the only thing this company sells.

What is it actually looking at?

A matrix of 755 machines from 128 makers, 709 of them checked against the maker's own specification table with the date we checked. A bench of 340 integrators with the applications and brands they build, which it reads but never names aloud. A casebook of published deployments. A library of peer reviewed research it may cite only when the search returns the paper in the same breath. When you say a 30 pound case at 90 an hour, it is not improvising. It is looking up what carries 14 kilograms at that rate and telling you which candidates do, with their published reach and rating, as candidates.

What is it forbidden to say?

That a configuration is safe. That a price is a price rather than a planning range. That a machine is eligible for your federal grant. That a study exists, unless the library just returned it. That it is a person. Ask it directly and it tells you it is an AI, and it will open the pod bay doors. It also will not help with layoffs, financing terms or taxes, and if you mention an injury it stops selling and tells you to call the people you should call.

Why trust a machine that is trying to sell you a machine?

Because it is built to lose sales. The company's margin is in the projects that go in and stay in, so the concierge is told to say no when the station is not the constraint, when the peak rate is out of reach, when parts arrive in a heap, or when a fixture would do the job for a tenth of the money. It is also multi-brand by construction. It ranks on fit first and discloses, in the same sentence, when a candidate happens to be a line we sell. A salesperson with one brand cannot make that promise. Software with 128 can, and does.

Where does it hand you to a human?

At three points. When the shortlist is real and the next step is confirming current specifications, availability and price with the makers. When the job needs a certified integrator and the team has to confirm that firm's interest and capacity before an introduction. And whenever you simply ask for a person, which it offers more than once because some people would rather talk than type. The AI's job is the first honest answer. The humans do the quote, the introduction and the walk across your floor.

The Dave problem

The famous failure of the fictional computer was not malice. It was a machine given two goals that conflicted and nobody to tell. The concierge has one goal, written plainly: tell the visitor the truth about their station, and sell equipment only where the truth supports it. It has a second thing HAL never had, which is a transcript emailed to the people who run the company after every real conversation, read by humans, used to fix it when it is wrong.

Try it with a job you already know the answer to. If it tells you something you know is false, we want the transcript. If it tells you something you did not know, that is the point.

Related on this site

Joe can tell you the class of machine, the published numbers, the disqualifiers and the honest range. He cannot tell you a cell is safe, cannot quote a price, and cannot see your floor. The humans do those three.

Put it to the test on your own job

Tell Joe what you are running and he will apply this to your numbers, including telling you when a robot is the wrong answer.

More notes