The short answer
Claude for Manufacturing, launched by Anthropic on 12 November 2026, gives manufacturers ready-made AI skills for quoting, quality investigations, breakdown diagnosis, supplier sourcing and production reporting. The tools are ready. The value comes from embedding them inside day-to-day operations so they run autonomously and safely, with guardrails around them and the business’s own rules built in.
What Claude for Manufacturing is
On 12 November 2026 Anthropic announced Claude for Manufacturing, a plugin for businesses on Claude’s Team or Enterprise plans. It adds 17 skills, which are written procedures Claude follows for a particular job, and four connectors, which let Claude work from data in Infor, Oracle NetSuite, Autodesk Fusion and Microsoft 365.
The skills cover sales, quality, operations, supply chain, design and setup. Each one ends with a person reviewing the work, and Claude can link the figures in a draft back to the documents and data they came from.
The six jobs it is built for
| Job | What Claude does | What it needs from your business |
|---|---|---|
| Quoting a new job | Lists what the customer’s request is missing, compares costs on similar past jobs and drafts the quote | Past orders, job costs and production steps |
| Investigating a failed inspection | Writes up the failure, searches past reports for the same problem, ranks likely causes and drafts a corrective action report | Inspection records and past quality reports |
| Diagnosing a breakdown | Reads sensor exports, repair notes, photos and the manual, and ranks likely causes for the technician to check | Machine data and maintenance history |
| Sourcing from another supplier | Prepares the request for quotes and compares the bids that come back | Supplier records and purchase history |
| Comparing designs | Generates variations of an existing Autodesk Fusion model, scores them against the engineer’s targets and drafts a decision memo | Autodesk Fusion |
| Reporting on production | Produces one report of output against target, equipment downtime, and what the downtime and missed output cost | Production, downtime and cost figures |
These are jobs most manufacturers will recognise, and most of them start with someone pulling information out of two or three systems by hand. In a Manufacturing Leadership Council survey published in September 2025, 79% of manufacturers said manually entered data was one of the sources of their manufacturing data.
The tools are ready. The work is embedding them.
Used on its own, Claude for Manufacturing is a capable assistant: someone opens it, asks for a quote or a report, and checks what comes back. That is a useful start, and it saves time.
The bigger gain comes when the same work runs inside your operations without anyone having to start it. A customer’s request arrives and a draft quote is waiting for the estimator. An inspection fails and the corrective action report is already written up. The production report is ready before the Monday meeting.
Getting there takes three things a ready-made skill cannot know about your business:
- Your systems. The work has to run on live data from the ERP, production and quality systems you use, whichever ones they are, so it is never working from last week’s export.
- Your business rules. Every manufacturer has rules that decide whether something can go out: credit limits, minimum order quantities, approved suppliers, margin thresholds, customer-specific terms. They need to be written into the process, not left for a person to remember.
- Guardrails. Clear limits on what runs on its own and what waits for a person, checks on every output before it is used, and a record of what was done and why.
Guardrails make autonomy safe
Autonomy is earned step by step. A process usually starts with every output reviewed. As the checks prove reliable against real work, routine cases are allowed through on their own and only the exceptions go to a person.
The guardrails are what make that safe:
- each output is checked against your business rules before it is used
- anything outside the rules, or that Claude is unsure of, goes to a named person
- actions are limited to what that process is allowed to do in each system
- every decision is logged, so you can see what happened and why
Where AI engineering comes in
Building that layer is AI engineering: connecting the tools to your systems, writing your business rules into the process, and putting the guardrails, testing and monitoring around it so it can run every day.
It is the work we do with manufacturers. For one UK manufacturer, emailed orders in any format are read, checked against more than 50 business rules and processed into their ERP, around 30,000 orders a year. You can see more examples in our AI agents for Business Central and our work with UK manufacturers and distributors.
What we have already built for these jobs
Several of the jobs Anthropic describes are ones we have already built and run for clients, embedded in their operations:
- Quoting. Enquiries are read from email or file, matched to the customer’s catalogue, priced per customer and issued as quotes and revisions. Tender and pre-qualification answers are drafted from past bids for the team to review.
- Reporting. An AI-powered data warehouse brings years of trading data from every source into one place, with an AI analyst the sales and leadership teams can question. Figures that disagree across systems are flagged, and each manager gets a daily summary of what has moved and what needs them.
- Sourcing and procurement. The agentic AI platform behind APSIS covers procurement and evaluation for construction teams, and supplier documents are checked against the order, the delivery and the budget as soon as they arrive.
Alongside these, we run the back-office work that feeds them: emailed orders processed into the ERP, supplier invoices and remittances matched, and "where is my order?" emails answered. Everything we have built is on our case studies page.
Where to start
- Pick one job. Choose the one that costs your team the most hours each week.
- Find where its information lives. List the systems and spreadsheets the job draws on.
- Write down the rules. Capture what a person checks before that work goes out today.
- Decide what runs on its own. Agree which cases need a person, at least to begin with.
- Measure the hours now. Then you will know whether it is paying back.
An AI readiness assessment works through these with you and puts the jobs in order of payback. Our AI in manufacturing guide covers the wider picture.
Frequently asked questions
Does Claude for Manufacturing work with our ERP?
It has connectors for Infor and Oracle NetSuite at launch, plus Autodesk Fusion and Microsoft 365. Other ERPs, such as Business Central, Sage, SAP Business One or Epicor, can be connected as part of embedding it in your operations.
Do we need Claude Enterprise?
The plugin is available on Claude Team and Enterprise plans, and an administrator installs it from the plugin directory.
Can it run without a person checking every output?
Out of the box, each skill ends with a person reviewing the work. Running parts of a process on their own is possible once your business rules and guardrails are built in and tested against real work, with exceptions still going to your team.
