Celonis's Shah Urges 3-Layer Context AI for Enterprise Decisions

1 articles · Updated · MIT Technology Review · Oct 22

Devavrat Shah said enterprise AI needs an operational context layer to make useful decisions, arguing agents without it can miss constraints, make costly errors, or waste computing power.
The 3-layer setup he described links systems of record and workflow logs below to AI agents above, giving models process awareness, guardrails, and outputs that humans can review before automation.
Shah urged companies to start with 1 specific business decision, target its hardest step, and measure gains in speed and value before scaling broader deployments.
He said the approach could automate repetitive and coordination-heavy work while shifting employees toward more strategic decisions, making workforce education and governance critical.
Looking ahead, Shah said context-aware AI could shorten planning cycles across finance, workforce management, marketing, sales, and other enterprise-wide decisions.