On November 12, 2026, Anthropic released Claude for Manufacturing, the first major industry-specific plugin for the Claude model, built for production teams. It includes 17 ready-made "skills" for factory tasks: from calculating new orders and investigating inspection failures to finding suppliers and generating production reports. The plugin connects to Autodesk Fusion, Infor, Microsoft 365, and Oracle NetSuite and works with records the factory already maintains. Every document draft must receive human approval before being sent to clients or suppliers. The plugin does not interact with physical equipment—Anthropic is only preparing the Model Hardware Standard, a standard for describing measurements and safety limits for programmable hardware.

image
image

What happened

The Claude for Manufacturing plugin was released on November 12, 2026. It is built around 17 built-in "skills" for typical production team tasks: calculating new orders, investigating inspection failures, diagnosing breakdowns, finding suppliers, comparing designs, and production reporting. Through connectors, the plugin connects to four enterprise systems—Autodesk Fusion (CAD), Infor (ERP), Microsoft 365, and Oracle NetSuite—and relies on records the factory already maintains, so no new infrastructure needs to be set up. A key operational condition is human-in-the-loop: every prepared draft, whether it is a commercial proposal, a corrective action report, or a procurement plan, must undergo human review and approval before being sent to clients or suppliers. The plugin does not control physical equipment. It can be installed on Claude Team and Enterprise plans.

Context

The release is not a new model or training method: it is a product layer over an agentic pattern, where industry skills rely on connectors to enterprise systems, and documents do not leave the perimeter without human control. The problem it addresses lives in the daily routine of factories: according to a 2025 survey by the Manufacturing Leadership Council, 79% of manufacturers still enter data manually, and transferring numbers between ERP, PDM, and email, as well as rewriting reports, remain the norm. The plugin takes on this part of internal labor, known in the industry as internal toil. An important architectural detail: the connector to the Infor ERP system provides read-only access, while write operations are moved to a separate partner plugin from Infor—this isolation of writes reduces the cost of potential model errors. In parallel, Anthropic is developing the Model Hardware Standard in research preview status: this is a standard for describing measurements, settings, and safety limits for programmable equipment, laying the groundwork for agents that will work with factory hardware in the future.

Why this matters for the industry

For the AI industry, this is a platform shift: Claude is being moved from chat to production routines for the first time, and integration is secured at the ERP level through a partnership with Infor, a company led by CEO Kevin Samuelson. Mandatory human approval of drafts and the absence of direct equipment control lower adoption barriers in regulated manufacturing, where the cost of error is high. For builders, the combination of "industry skills plus read-only connectors plus human control" reads as a reproducible template for building enterprise agents; it opens a niche for startups in skills and connectors to industrial systems, while simultaneously increasing the risk of platform dependency if the channel to customer data remains with the model vendor. The expected, but not yet realized, continuation is a wave of similar vertical plugins from ERP and MES vendors and their partners, as well as an expansion of the connector list.

Why this matters for users

If you work in manufacturing, logistics, or a related engineering role, the pilot is available on already paid Claude Team or Enterprise plans, and the data the plugin works with is usually already available at the factory. Teams whose records are in the connected systems will notice the effect fastest: skills cover the path from order calculation and failed inspection analysis to the morning production report; others will first need to organize their data. Drafts do not go to customers without your approval, so the first steps are safe for relationships with clients and suppliers. For observers, the release is useful as an indicator of change: LLM agents are moving from chat to multi-step workflows with access to enterprise systems.

What is still unknown / limitations

The sources contain no cases with numbers, benchmarks, or reliability metrics for the skills, so the claim that ready-made skills actually save time in production is currently unproven—this is an interpretation, not a measured fact. The details of control are also unknown: what exactly the reviewing human sees and how their approval is recorded in the system. The economics of the pilot—cost and delays—cannot be calculated until deployment figures appear. The market echo of the launch is currently zero: the corresponding post on Hacker News received one point and no comments. The Model Hardware Standard exists only in research preview status, and whether it will become an industry standard is an open question.

Sources

Author

Look at AI, editorial team