The next step in factory automation: Machines that adapt with AI

There are nearly 5 million industrial robots operating in the world's factories today, a number that grew nearly 10% in a single year. Almost all of them execute instructions written in advance by someone who anticipated the conditions they would meet. That approach has defined industrial automation for six decades.

Physical AI is moving intelligence into the work of manufacturing itself. Robots, vehicles, production systems, and connected infrastructure can now perceive changing conditions, reason over context, and act within defined boundaries.

Automation delivered precision, productivity, and safety by executing known instructions. Physical AI extends that foundation by helping systems respond to more variation in the conditions around them.

At the International Manufacturing Technology Show (IMTS) 2026, standing on the main stage in Chicago, Illinois, I framed the opportunity around a practical question: how do manufacturers turn intelligence that works in one machine or process into a capability they can deploy, govern, and improve across the enterprise?

Physical AI closes the loop between intelligence and action

Physical AI is intelligence operating in the physical world through machines, systems, and people. It works through a continuous loop:

1. Perceive: Read machines, environments, and operational signals.
2. Reason: Interpret that context and determine the next best action.
3. Act: Execute through machines, systems, and people, then use the result as feedback for the next cycle.

Connected operations gave people visibility. Industrial AI added reasoning over patterns and outcomes. Physical AI closes the loop by acting in the physical world, measuring the result, and improving the next cycle.

When AI moves from recommending on a screen to acting in the physical world, value and risk both become physical. Operating limits, identity and security, observability, validated fallback, and human approval and override must be designed into the system.

Physical AI is creating value across the industrial lifecycle

Manufacturers are already applying these capabilities from engineering through live operations.

At the engineering end, KUKA is making industrial robots easier to program and deploy. With iiQWorks.Copilot, users can describe a task, generate code, simulate the workflow, and deploy it. For simple tasks, programming can be up to 80% faster. Krones is using AI agents and physically accurate digital twins to speed filling simulations. Runs that took three or four hours can now take five minutes or less.

In live operations, ARUM is bringing the same shift into the machining center. TTMC Brain turns a conversation into a machining program and reduces numerical control programming from 177 steps to two. ABB is applying live industrial data to predictive maintenance and real-time optimization. Across its customer base, ABB reports up to 20% higher critical-asset reliability and up to 60% less unplanned downtime.

These examples address two constraints on scaling automation: the engineering effort needed to deploy it and the reliability needed to keep it running.

What physical AI changes for operations leaders

In my conversations with manufacturers, the challenge is increasingly orchestration. A successful machine or workcell is only the first step. Leaders have to connect that capability to the way the enterprise operates, decides, and scales.

Physical intelligence changes three enterprise models:

• The operating model moves from site automation to network coordination. Lines, plants, fleets, and field environments can operate as one connected system.
• The decision model moves from manual planning to agent-assisted operations. People set intent, policy, approval, and override while agents interpret context and coordinate work within those boundaries.
• The technology model moves from disconnected tools to one governed environment. Data, controls, platforms, and cross-vendor systems must work together across cloud and edge.

Those shifts depend on two connected loops. One accelerates physical systems innovation: teams simulate conditions, train and adapt models, validate behavior and failure modes, and iterate before deployment. The other operationalizes physical intelligence: teams orchestrate work, execute across machines and people, govern what happens, and optimize performance across systems and fleets.

Agentic orchestration connects the two. Operational feedback returns real-world conditions, interventions, failures, and outcomes to the next simulation and validation cycle.

Repeatability matters more than any single deployment. A successful machine or workcell becomes an enterprise capability when its data, integration, governance, and operating model can be reused across lines, plants, and fleets. Otherwise, the organization scales exceptions instead of capability.

AI governance in manufacturing belongs in the operating architecture

Identity, policy, monitoring, security, and accountability must travel with the capability from one machine or site to the next. Operational, engineering, and business context must inform the same decisions across cloud and edge.

For global manufacturers, that architecture must also support sovereignty, data residency, and operational technology security across sites and regions. Leaders need visibility into what the system can do, why it acted, and when a person must approve, intervene, or override.

A consistent trust model lets teams deploy, manage, observe, and improve capabilities across mixed environments without rebuilding the controls at every site.

For more on Microsoft's industrial AI platform, read Microsoft named a Leader in the 2026 Gartner® Magic Quadrant™ for Global Industrial AIoT Platforms.

Start with one bounded operation

Manufacturers do not need to automate an entire plant to begin. A practical starting point is one operational problem where the outcome matters and the boundaries are clear.

Choose a workflow with measurable value, such as quality inspection, maintenance triage, machine programming, or process optimization. Map the data and operating context required to make a sound decision. Define what the system may recommend, what it may execute, and where a person must approve or intervene. Then design the deployment so that the successful pattern can be reused across another line, site, or fleet.

For examples of how manufacturers can scale high-value use cases, read Step into the Factory of the Future.

Physical AI will be measured in the operation: faster engineering, safer work, more reliable assets, and systems that improve under clear human authority. Start with one consequential workflow. Define its boundaries. Build so success can repeat.