Can a shop my size afford flexible automation if the production mix changes every week?

Not from this announcement. The deployments described are consumer electronics lines at volume, and the IFR still names high initial investment and total cost of ownership as the barrier for smaller manufacturers. What you can afford this month is the measurement: changeovers per year, line hours lost per changeover, and scrap in the first hour after restart. That number decides everything later.

Is this different from a collaborative robot I can buy today?

The claim is different. A cobot still runs a programmed motion, so a new part usually means new programming or new fixturing. Industrial Next's claim is that a cell taught by demonstration and simulation keeps running on a part and a line it has never seen, at or above human speed with 100% accuracy on a film peeling station. That claim is from the company, not from an independent test.

What should I ask a vendor before signing anything?

Three things in writing: the acceptance test (cycle time, accuracy, and what happens to a bad part), the cost and downtime of switching the cell to a new part number, and who owns maintenance and safety assessment. If the switching cost is not in writing, you are buying a programmed cell with better marketing.

What task-switching robots change for a small shop, and what they do not

The cost that keeps a small shop out of automation

A traditional industrial robot is programmed for one specific repeatable task, so a small change in the part or the process either makes it fail or buys expensive downtime, which pushes manufacturers to simplify products to suit the machine (TechCrunch). For a shop running short, frequent changes that is the whole problem: you pay for the cell once and for the reprogramming every time the mix moves.

The money is already in the room. The IFR puts the market value of industrial robot installations at an all-time high of $16.5 billion with about 4 million industrial robots running in factories, and still names high initial investment and total cost of ownership as what holds smaller manufacturers back (The Robot Report).

What was actually announced

Industrial Next, founded in 2021 by two engineers who previously worked at Tesla, builds cells trained by demonstration and a high fidelity simulation engine rather than programmed motion by motion. On stage it said BMW Group Technology Office will evaluate its dexterous manipulation for manufacturing. The example given: a film peeling station that strips protective film off camera modules and buttons at or above human speed with 100% accuracy, including after being moved to a phone model and a production line it had never seen (TechCrunch).

Those are the company's own claims from a conference stage. I have not seen an independent audit of them, and neither has anyone else writing about this.

Work the number on your own line

A vendor conversation is only worth your time once you know this:

changeover cost per year = changeovers per year x line hours lost per changeover x your loaded cost per line hour

Then add two inputs nobody asks for: how many distinct part numbers a single station handles in a month, and how much of your scrap happens in the first hour after a restart. The second one is usually the ugly one, and it is exactly the part a cell that holds accuracy across parts would fix.

Compare that annual number against three years of owning or renting the cell. If it is a rounding error, the answer is a better fixture, not a robot.

A six-line qualification checklist

Before you take any vendor call:

One station, one task, and the task is dull, two-handed and currently done by someone you cannot hire.
The part changes often, the motion does not: pick, place, peel, thread, snap.
Tolerance is millimetres, not microns.
You can record about twenty to forty clean demonstrations of the task, done by a person who does it well.
You can write the acceptance test yourself: cycle time, accuracy, and what happens to a bad part.
The vendor will state the cost of switching that cell to a new part, in writing, before you sign.

Four or more no answers and this is not your project yet.

Limitations

Every figure above is a vendor claim from a stage or an industry association estimate. None of it was measured on a shop like yours.
The described work is consumer electronics at volume. A fabricator or machine shop has heavier parts, a dirtier floor and far fewer units per part number.
Demonstration training does not shorten safety assessment, guarding, part presentation, fixturing or finding someone to own maintenance.
Accuracy at one station is not uptime for a line.
My guess: for a small shop the realistic route in the next year or two is renting a cell rather than buying one, since the IFR names robotics-as-a-service as the path with no fixed capital involved. I have not tested a provider.

My take

The headline number is minutes instead of weeks. The number that matters is zero: zero reprogramming when the part changes. That is what prices short-run manufacturers out of automation, and if demonstration-trained cells really hold accuracy on parts they have never seen, flexible automation stops being a volume game.

I would not sign anything this quarter. I would walk the floor with a stopwatch, because the shop that already knows its changeover cost is the one that can say yes quickly when this becomes buyable. More on where automation pays inside a small business on the blog.