What came out on November 12
Anthropic launched Claude for Manufacturing, a plugin designed for people who work in a plant: sales, quality, maintenance, purchasing, engineering, production. Inside are 17 skills, meaning prewritten procedures that tell Claude how to handle a typical factory task, and four connectors to the systems it reads data from.
It is available on Team and Enterprise plans. You install the plugin, tell Claude "Set us up" and it walks you through configuration, flagging which skills can't yet run in full because a piece of data or a connection is missing.
The launch lands in the middle of the roadshow for Anthropic's stock market listing, and it is a signal: after finance, legal and software development, Anthropic is looking at industry with a dedicated product. There is a technical webinar on November 17 at 11:00 ET.
The skills that matter
Anthropic has not published the full list of 17 skills, but it described the main ones, and they are the things that take up more time in a plant than they should.
In sales, quoting a new job: Claude reads the customer's request, lists the missing information, compares costs with similar past jobs and drafts the offer.
In quality, analyzing a failed inspection: comparison with past reports, likely causes ranked by probability, a draft corrective action. The acronyms are the ones quality people know: NCRs for nonconformances, 8D for structured problem solving.
In maintenance, diagnosing a fault from sensor data, repair notes, photos and the machine manual. In purchasing, requesting quotes from a new supplier and comparing responses. In engineering, variants generated from an existing model in Autodesk Fusion. In production, a report on output against target, downtime and the cost of lost production.
The connectors, and the European problem
There are four connected systems. Autodesk Fusion for design. Infor to look up purchase orders and inventory. Microsoft 365 for procedures, supplier emails and reports in shared folders. Oracle NetSuite for prices, inventory, production steps and job costs.
With Infor there is a distinction to understand. The connector included in the plugin only reads. To change data, for example releasing a purchase order or rescheduling production, you need a separate Infor product, available to some Velocity Suite customers with their own Claude licenses, and every change requires the user's approval.
The problem for many European manufacturers is obvious. In small and mid-sized plants NetSuite and Infor are not the norm. People work on SAP or on local and vertical ERPs. For systems without a connector Anthropic suggests periodic exports saved to Microsoft 365, which is practical but limited: Claude works on a snapshot of the data, not live data.
The more solid route is to build the connection to your own ERP with an MCP server. I wrote about it for SAP.
Get updates on Claude and AI for business
One email when there's something worth reading. No spam.
Evaluating Claude for your organization? Learn what it costs or which plan fits best
What it doesn't do, and Anthropic says so
The clearest sentence in the announcement is this: Claude for Manufacturing doesn't operate equipment. It doesn't control plants, doesn't talk to PLCs, doesn't intervene on the line.
For that side Anthropic is working on something else, the Model Hardware Standard, now in research preview, designed to let AI operate physical equipment safely. It is a separate project, far from production.
The other limit is deliberate: every skill ends with review by the person responsible. The sales lead approves the quote, the quality manager signs off the corrective action. Claude can link every figure and claim to the document it came from, which makes review faster, but it doesn't remove it.
I think that's the right approach. In a plant an error in a quote or a root cause analysis costs real money, and a tool that prepares the work and has it signed by whoever is accountable is easier to get accepted than one that promises to do it all alone.
The case that shows where this is heading
Alongside the plugin Anthropic described an experiment with Diode, an electronics board maker. With Claude Fable 5.2 three complete devices were designed: a USB-C speakerphone, a voice assistant hub and a small handheld gadget. The boards were manufacturing-ready in under a day and all three worked without changes to the circuit.
It is a showcase, chosen by the company selling the model, and should be taken as such. But it says something about direction: assisted design no longer stops at the drawing, it reaches a working part.
For most manufacturers the immediate value isn't there. It is in the two hours a day the quality manager spends reconstructing the history of a defect, or the three days it takes to answer a complex request for quotation. I wrote about the model used in the experiment in Claude Fable 5.2.
How I'd use it in a plant
I'd start with one area, the one where data already lives in documents and not only in the ERP. Quality is almost always the best candidate: nonconformance reports, 8Ds, technical sheets, customer complaints. They are text, they already sit in shared files, and searching through past cases is exactly what Claude does well.
Second step, quotes, but only if the job history is accessible in an orderly way. Without comparison to similar jobs the skill does half the work.
Third, maintenance, once there is a minimum of sensor data and the machine manuals are digitized.
One figure helps explain why it's worth it. According to a 2025 Manufacturing Leadership Council survey cited by Anthropic, 79% of manufacturers still enter data by hand. The bottleneck isn't the model's intelligence, it's the distance between the data and the people who need it. For the broader picture there is our article on Claude for manufacturing.