The arc
I did not enter Industrial AI from a generic AI background. I arrived through asset operations, maintenance, mobile work, enterprise systems, infrastructure analytics, IoT, geospatial and environmental context, predictive modelling and asset investment planning.
The eras
OperationsArmy procurement, workshops and asset lifecycle responsibility.
EAM + work executionPinnacle, TechnologyOne, IBM Maximo, ABB Ellipse and IFS, including mobile work and field-service design.
IoT + analyticsIBM Smarter Buildings / Cities and connected operational intelligence.
Contextual asset analyticsYotta Horizons: condition, geospatial, environmental and demographic context feeding predictive infrastructure and investment decisions.
AIPABB investment planning and PowerPlan / Riva.
Agentic AIModern work around context, autonomous workflows, confidence and governed action.
Patterns that travel across Industrial AI
The recent products sit in different domains, but the recurring architecture is surprisingly consistent. The useful question is not which app an industrial problem resembles. It is which decision pattern is being solved.
Context before reasoningSymphony: assemble the evidence around a decision before asking AI to recommend anything.
Goals with bounded authorityGNR8: give an agent an objective, constraints, resources, approval points and connected actions.
Ask when evidence is missingPhoenix: detect gaps explicitly and escalate rather than allowing plausible invention.
Monitor the dependency marketOffice Cockpit: know which capability depends on which model and what happens when that dependency changes.
Industrial AI is not one model bolted onto an ERP screen. It is a set of decision and action loops built around operational context.
What AI changes
The opportunity is no longer simply to report what happened or predict what might happen. AI can assemble context around an event, reason across structured and unstructured information, recommend an intervention and increasingly initiate action in the enterprise systems that execute the work.
That makes the old asset-management problem more interesting, not less: how do we connect condition → risk → decision → investment → execution while keeping the evidence chain intact?