FANUC's AI welding agent: should you wait to buy a cobot?

No. FANUC's AI Welding Agent, announced on 11 September and shipping from the end of December, is a subscription that runs on hardware you would be buying anyway: a CRX cobot and the tablet teach pendant it already comes with. Nothing in the cell changes — arm, torch, power source, fixture table, bottle — so a welding cell that pays for itself today still pays for itself, and one that does not will not be rescued by a software line FANUC has not put a price on.

Buy or outsource on what the cell does now. If the agent turns out to save the hour your welder currently spends teaching each new fixture, that is upside on a machine you already justified.

What the agent does, and what it leaves you

The welder photographs an engineering drawing with the camera built into the CRX tablet teach pendant. The agent reads it, works out the material and the finished shape, and generates the welding parameters — current, voltage — and the robot motion program. FANUC calls that zero setup and zero teaching. No extra camera or box is needed at the cell, it works with any welding power supply connected to a FANUC robot, and it is subscribed through your existing FANUC arrangements rather than a separate Google Cloud contract. FANUC also states that drawings and production data processed by the agent are not used to train other users' models.

What it does not do is sign anything off:

"Operators can either execute the welding parameters and robot motions generated by the AI Welding Agent as-is, or fine-tune them before welding." — FANUC Corporation, 11 September 2026

FANUC's stated reason for building it is the shortage of welders who can optimise parameters and teach robot motions. So the person checking the agent's output is the person you are already short of. Read the saving as an hour off each new part, not a welder off the payroll.

The release carries no price, no cycle time, no programming-time figure, and no list of the joints, materials or thicknesses it covers. Until those exist there is nothing to put in a capital case.

What to specify if you buy the cell now

Four things, and only one of them is about the agent.

• A CRX with the tablet teach pendant. The camera the agent uses is in that pendant. A cell built around anything else cannot pick the agent up later, and that is the one decision made now that is hard to undo.
• Reach before payload. A CRX-10iA carries 10 kg to 1,249 mm; the /L carries the same to 1,418 mm, at ±0.05 mm repeatability against ±0.04 mm for the shorter arm. A torch and cable pack weigh little, so the fixture envelope decides the model, not the load.
• The power source on its own merits. The agent is independent of it, so choose on wire feed, duty cycle and who services it.
• Your real programming time, at today's method. Lead the arm to the start of the seam, record it, lead it to the end, record that, set the weave, the travel speed and the wire feed, dry run, weld. Cost that honestly. It is the number the agent will later be measured against.

One more thing worth a hard look before automating from drawings: the drawings. The agent reads what you show it. A shop where weld symbols land on the wrong side, throats go missing and revisions never reach the floor is automating from a bad input, and nothing in the release addresses a drawing that does not add up. Seam tracking corrects the path to the part. Nothing corrects the print.

If the welded assemblies are needed before any of this carries a price, they do not have to wait: Vesprr's workshop runs MIG, TIG and arc welding for structural steel and custom assemblies.

Sources

• FANUC Corporation, 11 September 2026: FANUC Accelerates Physical AI in Arc Welding with the New "AI Welding Agent" — the capability description, the operator sentence, the deployment and security statements, and the end-of-December shipment date.
• FANUC Corporation: FANUC Robot CRX series — payload, reach and repeatability by model.
• Video: Easy to Use CRX Cobots for Welding — FANUC America Corporation.