Anthropic Ships Claude for Manufacturing: 17 Skills, Four Connectors, and a Rule That the AI Never Touches the Machines

Anthropic has released Claude for Manufacturing, its first industry-packaged plugin, available to Team and Enterprise customers from November 12. The move takes Claude out of the general office suite and into the specific paperwork of running a factory: pricing jobs, chasing quality escapes, diagnosing stopped machines, sourcing suppliers, and assembling production reports.

The plugin ships with 17 skills organized across six areas of plant work — sales, quality, operations, supply chain, design, and setup. Each skill follows written procedures and hands its output to employees for review. The workflows read like a factory's daily grind: drafting a quote from a new inquiry, summarizing failed inspections into candidate causes and corrective actions, building a morning production brief from output, downtime and scrap data, turning a breakdown into a shift-handover report, and lining up alternative sources when a supplier's parts arrive out of spec.

Connectivity comes through four connectors: Autodesk Fusion, Infor, Microsoft 365, and Oracle NetSuite. The idea is that Claude works from a manufacturer's existing records — design files, purchase orders, inventory, job costs and procedure documents — instead of asking people to re-key data. Anthropic cites a 2025 Manufacturing Leadership Council survey in which 79 percent of manufacturers still manually enter data between tools that do not connect, which is the problem the plugin is built around.

The limits are as deliberate as the features. Anthropic is explicit that the plugin does not operate equipment: a breakdown diagnosis produces likely causes and a checklist for a technician, not a repair order. And it does not write to systems of record — the Infor connector inside Claude for Manufacturing is limited to record lookups. Infor, which launched its own Infor Industry AI for Claude the same day, offers the deeper integration: actions like ordering from backup suppliers or rescheduling production, but only after human approval.

The design shows where Anthropic thinks the enterprise agent market is going. Rather than promising autonomous factories, the plugin positions Claude as the layer that drafts, cross-references and assembles — with every output linked back to source documents so a quality engineer can check the reasoning before it reaches a customer or supplier. A separate research effort, the Model Hardware Standard, is exploring standardized drivers to connect AI agents to physical devices like robotic arms, but that remains a research preview precisely because models still have a limited grasp of physical processes.

For Anthropic, the manufacturing vertical is a bet that industry-specific packaging — skills, connectors, guardrails — will matter more than raw model scores as enterprise buyers start deploying agents into regulated, physical-world businesses. A technical webinar on November 17 will demonstrate the plugin. The competition, notably, is already there: Microsoft has been pushing industry agents through Copilot, and vertical AI startups have been quietly signing up manufacturers for years. Whether a cross-industry plugin beats an embedded vertical vendor is the question the next few quarters will answer.