1. News
  2. CRM, SaaS & automation
  3. Microsoft · Robotics
Microsoft · Robotics2 min read

Microsoft Pushes AI That Lets Factory Robots Think on the Job

Microsoft is promoting physical AI, software that lets factory machines sense, decide and act on their own. Small manufacturers are part of the pitch.

By the Dashline deskSource: Small Business Trends
Microsoft Pushes AI That Lets Factory Robots Think on the Job
Photo: Enric, CC BY-SA 4.0

Microsoft is pushing what it calls physical AI, a way for robots and factory machines to read what is happening around them and decide what to do next, instead of repeating fixed steps. The idea is aimed at manufacturers of all sizes, including small shops and plants that want less downtime and lower costs.

What changed

Nearly five million industrial robots today follow pre-set instructions. Physical AI is meant to let machines interact with their surroundings and work through changes and surprises in real time. The bigger goal is to take intelligence that works in one machine and make it usable across a whole company.

Michael D. J. O’Connor, a speaker at the International Manufacturing Technology Show (IMTS) 2026, put the question this way: "How do manufacturers turn intelligence that works in one machine or process into a capability they can deploy, govern, and improve across the enterprise?"

How it works

The approach is a three-step loop that repeats.

  • Perceive: the system reads signals from machines and the environment.
  • Reason: it interprets that data and picks the best course of action.
  • Act: it carries out the plan through machines or people, then learns from the result.

Early numbers

Two industrial companies are cited as examples. KUKA says robot programming can be up to 80% faster for simple tasks, because users can describe a job and have code generated. ABB reports up to 20% higher reliability in key equipment by using data for predictive maintenance, which means spotting problems before a machine fails. These are company-reported figures.

What to do

The advice is to start small. Pick one bounded task with results you can measure, such as machine programming or quality inspection. Then decide whether the AI should only recommend actions, carry them out, or work under human supervision.

Governance matters too. Businesses need rules for identity, security and accountability, a way to see how and why the system made a decision, and clear points where a person must approve.

What is not clear yet

The coverage does not say what Microsoft's offering costs, when it will be available, or which tools small manufacturers would use. It also does not say how much of the promised gain a small plant can expect. Those details would need to come from Microsoft's own announcement.

The takeaway

Test physical AI on one measurable task first, and decide in advance when a person has to approve what the system does.

More articles

You should really read these.

Next step

Let’s talk