Honeywell Technologies chief executive Vimal Kapur says industrial systems have become so complex that people can no longer keep up with them unaided. His proposed answer is “physical AI”: a learning layer over the automation already running places such as hospitals, data centres and refineries.
Watch Desk analysis
What happened
In an interview with Fortune, Kapur said some facilities run themselves “maybe 95%” of the time, with people handling exceptions. He argued that decades of added software and physical infrastructure have made these systems difficult for operators to oversee. Those figures and claims are Kapur’s, as reported by Fortune.
Kapur’s answer is not primarily humanoid robots. He described AI applied to large physical assets, using data held within individual buildings and facilities. Honeywell launched Forge Cognition, an AI platform developed with NVIDIA and Dell, in October; Fortune says its autonomous mode had not been fully switched on at the time of the interview. Read Fortune’s report.
The practical obstacle is access to that data and the expertise needed to make it useful. Kapur called data access the “first friction” and said domain knowledge is essential. His pitch is that AI could help workers manage complex systems, not simply replace them.
Why it matters
Adding an AI layer to industrial automation is a consequential step: it could help people make sense of systems that are already difficult to monitor, but it also raises the stakes when something goes wrong. As Kapur describes it, people remain responsible for exceptions, precisely when conditions may be most difficult.
That tension is not new. Fortune points to psychologist Lisanne Bainbridge’s 1983 paper, “Ironies of Automation”, which argued that operators can be left with the hardest tasks after routine work is automated. More capable systems do not make human oversight disappear; they may make the quality of that oversight more important.
Our read
There is a real problem here, from ageing infrastructure to shortages of skilled workers. But “physical AI” is not proof that a complex system has become safer or easier to run. Honeywell’s platform is a concrete example of the direction Kapur is advocating, while the report says its autonomous mode was not yet fully active. The useful test is whether these systems help operators handle exceptions, with clear evidence of what the AI can and cannot do.
What to watch
- Whether Forge Cognition’s autonomous mode is switched on and where it is deployed.
- How Honeywell explains the data access and domain expertise needed at each site.
- Whether operators report better handling of exceptions, rather than simply fewer routine tasks.
- How industrial AI changes workers’ responsibilities and reskilling needs.
Discussion spark: If AI takes over most routine operation of critical facilities, should the priority be making the systems more autonomous or making sure people can still handle the exceptions?
Sources and evidence
- Source update (16 November 2026, 22:04 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.