Planning
P&G runs 20 forecasting models a day to plan six weeks of demand
P&G evaluates 20 AI models every day to produce a six-week demand forecast for Charmin and Bounty, its Director of Demand Forecasting told an IIoT World panel. The detail worth noting is the plurality: an ensemble re-evaluated daily, not one model trusted continuously.
Model selection is an operating routine, not a one-time procurement decision. Ask a forecasting vendor how often their model is re-checked against alternatives.
Workforce
Operators say industrial AI stalls on everything except the technology
At Yokogawa’s YNOW2026 conference, operators, investors and startup founders agreed that industrial AI’s biggest obstacle is not the technology, Control Global reports, but everything around it — data access, ownership, and who is accountable for acting on an output.
Budget the change work, not just the licence. A pilot that stalls usually stalls on ownership, not on accuracy.
Agents
Agentic AI reaches most of the Fortune 500 and hits the integration wall
MIT Technology Review reports agentic AI has been adopted by some 80% of Fortune 500 companies, and that the live problem has moved from capability to plumbing: how agents work together, connect to the systems and data they need, and operate safely across a business workflow.
The hard part of an agent pilot is system access and safe failure, not the model. Scope integration before you scope use cases.
Deployment
Process historians cannot supply what physical AI needs from them
IIoT World argues that physical AI — systems reading sensor telemetry, predicting equipment behaviour and triggering automated responses — needs two things historians were never designed to give: high-resolution historical datasets for training, and continuous real-time telemetry for inference.
Check sample rate and retention on your historian before scoping a predictive project. The data layer is usually the schedule risk.
Frontier
NVIDIA plans to acquire Hugging Face, pledging to keep the platform open
NVIDIA has announced plans to acquire Hugging Face and says it will keep the AI development platform open, The Robot Report reports. NVIDIA is already a major model contributor to the platform, which recently surpassed one million datasets.
The open-model distribution layer would sit with your GPU vendor. Worth checking where your AI stack assumes neutrality between silicon and models.
