Physical AI is moving from servers to shop floors
AI has stepped out of IT and into operations. Anant Maheshwari, who leads global regions at Honeywell, put it plainly: factory systems that could "see, think, and act" now add one more verb-"learn."
That shift matters for how we run plants, buildings, and complex infrastructure. It changes who can operate assets, how quickly they ramp, and how we design supply chains.
From rules to learning loops
Traditional control systems were built on fixed rules set by engineers. Going forward, systems learn from data and create new rules as they work, under human oversight.
This shows up in real use cases: tighter process control, energy optimization in buildings, predictive maintenance that updates itself, and safety systems that adapt to context instead of firing on static thresholds.
- Factories: closed-loop quality control, computer vision on lines, adaptive scheduling tied to constraints.
- Buildings: occupancy-based HVAC, dynamic setpoints, automated fault detection with self-tuning.
- Infrastructure: anomaly detection on utilities, condition-based maintenance for fleets and assets.
Jobs: augmentation and faster talent maturity
There's a talent shortage across high-stakes industries like semiconductors and aerospace. You don't put a year-one operator on a critical asset without support.
AI can shorten time-to-proficiency. Think digital mentors that surface next-best actions, simulations that let teams practice failure modes, and guided checklists that learn from expert behavior. Enablement is the point.
- Capture tribal knowledge: convert SME steps, signals, and exceptions into machine-readable SOPs.
- Deploy copilots at the HMI: suggestions, alerts with context, and one-click playbooks.
- Use simulation first: sandbox new rules, then shift to shadow mode before live control.
- Keep guardrails: permissioning, setpoint limits, and instant human override.
What operations leaders can do this quarter
- Pick one high-variance process with clear ROI (yield, energy, throughput) and run a 60-90 day pilot.
- Add "learn" safely: start in advisory mode, then move to closed loop for low-risk parameters.
- Fix data plumbing: time-series alignment, sensor calibration, and event labeling.
- Stand up human-in-the-loop: require operator confirmation for rule changes.
- Measure what matters: OEE, MTBF, energy per unit, defect rate, recovery time.
- Secure it: apply least-privilege access, network segmentation, and backup procedures.
- Prepare people: training, clear role changes, and a feedback channel to product/engineering.
- Vendor checklist: integration with your SCADA/MES/BMS, on-prem/hybrid support, audit logs, and SLA on model drift.
Supply chains are going local-first
Global supply chains have shifted. As Maheshwari noted, more companies are moving to country-specific strategies. Decades ago, India and China already had dedicated plans; that model now fits more markets.
The lesson: build local capability-leadership with cultural nuance, suppliers with shorter lead times, and regional compliance baked in. The "India model" is a practical playbook many multinationals are now applying elsewhere.
- Create local innovation cells that ship features for the region, not just HQ requests.
- Dual-source critical components; keep a minimum viable local supply line.
- Empower local leaders with P&L and procurement authority.
- Standardize interfaces (APIs, data models) so global systems can plug into local stacks.
- Run quarterly stress tests on logistics, inventory buffers, and substitution rules.
Why India matters for ops
India is a benchmark right now-both a manufacturing base and a source of innovation. Honeywell continues to build and ship from India for global markets, which signals mature engineering, supply, and talent depth.
For operations teams, that means more options: co-develop solutions with India-based teams, source advanced controls locally, and use the region as a training and support hub.
Governance and risk (keep it boring and consistent)
Learning systems need structure. Align AI projects with a clear risk framework, enforce change control, and keep SMEs in the loop for any rule updates that touch safety, quality, or compliance.
For guidance on policy, roles, and controls, see the NIST AI Risk Management Framework here.
Quick checklist for your next AI-in-ops pilot
- Business case: 1 metric, 1 owner, 1 deadline.
- Data: labeled events, aligned timestamps, sensor health checks.
- Safety: limits, overrides, audit trails.
- People: training plan, SOP updates, escalation paths.
- Process: shadow mode, phased rollout, rollback plan.
- Review: weekly triage on drift, false positives, operator feedback.
Want structured upskilling for your ops team? Explore focused AI learning paths by job role here.
Your membership also unlocks:
