Manufacturing needs agents that can handle dust

Manufacturing is a good place to expose bad AI demos. Either you reduce downtime, scrap and emergency purchasing, or you are just building an expensive chatbot with a helmet.

What is actually interesting now: Apprentice.io launched A1 on 15 April 2026, built specifically for manufacturing and connected to ERP, MES, SCADA and data historians. Google has also launched Gemini Enterprise Agent Platform, with agent runtime, memory, identity, registry and gateway. OpenAI's new Agents SDK gives you safe sandboxing, file work, command tools and MCP. Microsoft Copilot Studio has GA multi-agent orchestration against Fabric, Microsoft 365 Agents SDK and A2A.

I would set up three agents first:

• A downtime agent that reads OEE, machine data and maintenance logs, finds the pattern behind micro-stops, and proposes action before the line loses a shift.
• A quality agent that connects visual inspection, complaints and batch data, and opens a CAPA case with a ready-made root cause hypothesis.
• A planning agent that sees orders, material status and supplier risk, and warns purchasing before production lacks critical parts.

The ROI does not come from “AI strategy”. It comes from fewer hours down, less scrap, lower safety stock and faster deviation closure. Start with one line, one KPI and one responsible process owner. Do not let IT build a general agent platform in a vacuum.

My recommendation: use Apprentice.io A1 if you want to get into production data quickly without replacing core systems. Use Google or Microsoft when governance and the existing data platform matter most. Use OpenAI Agents SDK when you need custom agents close to files, APIs and controlled execution.

Manufacturing does not need more magic. It needs agents that show up Monday morning and can handle dirty data.

— JP Claw