Celonis chief scientist: enterprise AI fails without operational context
MIT Tech Review AI · rss · 2026-10-22
In MIT Tech Review's Business Lab podcast, Celonis chief scientist of enterprise AI and MIT professor Devavrat Shah argues that enterprise AI's biggest gap in moving from experimentation to scale is operational context.
- Frontier models applying generalized principles to specialized scenarios either miss critical constraints (making costly mistakes) or burn massive compute figuring things out through repeated questioning.
- His proposed fix: a contextual layer connecting agents to the processes and data enterprises already use — Celonis delivers this via a real-time digital twin — so agents understand workflows, operate within guardrails, and produce reviewable recommendations.
- Practical advice: pick one specific business decision, identify its hardest steps, test whether AI improves them, then measure speed and value gains before scaling.
- Longer term, automating repetitive work could shift employees toward strategic decisions, but requires workforce education; opportunities extend to enterprise-wide planning in finance, workforce management, marketing, and sales.
Produced in partnership with Celonis.
More from Companies & People
- Nigeria AI strategy co-author Chinasa Okolo joins Build 2026 on Africa's shared compute — ChinasaTOkolo · 2026-10-22
- Nadella: TK Elevator uses AI agents to speed maintenance of 1M+ elevators — satyanadella · 2026-10-22
- Indie Dev Building Full-Stack AI Infra Primitives Says X's Gatekeeping Beat LinkedIn's Reach — ShimazuSystems · 2026-10-22
- CrowdStrike CEO: Adversaries Already Run AI Agents, Fastest Attacks Play Out in Seconds — No Priors · 2026-10-22
- Stratechery interviews Craig Moffett on SpaceX and lessons from the dot-com bubble — Stratechery · 2026-10-22
- Law professor rethinks Microsoft Copilot after hands-on work inside a major law firm — jkubicki · 2026-10-22