Long reads, with pictures

The next ten years, without the hype

October 2, 2026 · 4 min read

Everyone selling robots has a forecast, and most of them are a product launch wearing a trench coat. Here is ours: what a small plant should buy now, what it should watch, and what it should ignore for a while, with the reasoning stated so you can disagree with it. Our read, not a prophecy.

The short answer

The machines that will matter most to a small plant in 2036 are mostly the ones that matter now, made cheaper, easier to set up and better at seeing. The headline technologies, humanoids chiefly, will matter to somebody, and probably not to a sixty person job shop for most of the decade. This is our read, written the way the rest of this site is written: with the reasoning visible, no numbers we cannot source, and an invitation to argue. Forecasts are the one product this company sells with no warranty at all.

Buy now: the machines that already pay back

Collaborative arms on carts for machine tending. Palletizers on columns. Welding cobots on tables. Mobile robots with hooks and shelves. Autonomous scrubbers in buildings with enough floor. Force controlled finishing cells. Caged indoor drones, or the hired flight, for any vessel you scaffold. Every one has published deployments in this site's casebook at companies with two or three digit headcounts and planning ranges in its guides. None will be obsolete in ten years, because the job they do, taking the motion that hurts away from a person, will not be obsolete either. The honest forecast for these is boring: cheaper, with better software, and with more makers competing on fit.

Watch: three things maturing in plain sight

Vision. The 3D cameras and learned detection that made mixed-case depalletizing possible are getting cheaper every year, and the frontier is moving from finding a box to finding anything in a bin. A shop with random part presentation should watch this closely, because the fixture that solves presentation today may be unnecessary in five years. Buy the fixture anyway. It is cheap, and it helps the people in the meantime.

Docks. A drone that lives in a box on site and flies itself on a schedule turns inspection into monitoring, as the solar operator in the casebook found at utility scale. The hardware is here; the approvals to fly unattended are the slow part, and they belong with specialists. Watch for the day a dock on a plant roof is as ordinary as a security camera.

Navigation. Mobile robots that handle ramps, thresholds, outdoors and elevators without a project each time. Each year the building that disqualified an AMR last year gets smaller. If you were told no because of a ramp, ask again in two years.

Watch with care: the humanoids

They are in the news, in the matrix on this site, and not for sale here, because the dealer channel barely exists and almost nothing about them is published in a form a buyer can verify: duty cycles, service, parts, safety assessment under a standard that fits a walking machine. Some of that will change this decade. Some of it will take longer than the keynotes suggest. The honest advice for a small plant is that the general purpose robot has been a year away since the 1950s and the specific one is on a truck, and that advice has a very good record. When a certified integrator in your state will sign a risk assessment for one and a dealer will stock parts, it becomes a real option. Not before.

Watch with suspicion: anything sold by the forecast

A machine justified by what it will do in version three is a machine you are financing for the vendor. The guides on this site put the disqualifiers before the pitch for a reason. If a seller's case for a cell depends on a feature that is coming, price the cell without the feature and see if it still pencils. If it does not, wait.

The thing we are most sure of

That the honest first answer will keep getting cheaper and will change who automates. For a century the first conversation about a robot was with someone paid to sell one brand, or a feasibility study a small plant could not afford. That is the barrier that actually kept robots out of the sixty person plant, more than any limit of the machines. An AI that searches a neutral matrix, ranks on fit, discloses its interests and says no is a strange thing for a dealer to build, and we think it is the most durable forecast on this page: the plants that automate well in 2036 will be the ones that got a straight answer in 2026.

How to read any forecast, including this one

Ask what it costs the forecaster if they are wrong. Ours costs us the sale we talked you out of and the trust we need for the next one, which is why the hedges stay in. Then do what this site does with every number: put it next to a published account, check the source, and try it on your own station. The concierge is where that starts. Describe the job to Joe and ask him which of these three lists it belongs on.

Related on this site

• The FCC humanoid rule: what a plant can still buy
• The hundred-year conversation between people and machines
• Small shops, big machines: ten robots that fit in a corner
• Can an AI really tell you whether your job is automatable?
• The open robot and aircraft matrix