Enterprise AI needs operational context to move beyond experimentation, Celonis chief scientist argues
Without understanding business processes, AI agents risk costly mistakes and wasted compute, argues Devavrat Shah of Celonis in an MIT Technology Review interview.
1 source · cross-referenced
- Enterprise AI often fails to scale because agents lack operational context about how businesses actually operate.
- Devavrat Shah, Celonis chief scientist, says AI must understand workflows, constraints, and guardrails to deliver value.
- Companies risk costly errors or excessive compute use if AI acts without the right business context.
- Shah advises starting with specific decisions, measuring speed and value gains, and investing in workforce education.
Enterprise AI can generate recommendations, automate tasks, and accelerate decision-making, but without an understanding of how a business operates, even powerful models may miss critical constraints and relationships that determine whether a decision is correct, according to Devavrat Shah, chief scientist of enterprise AI at Celonis.
Shah argues that AI agents can support individual tasks within larger decision-making processes, but they require context to act usefully within complex business workflows. Without this operational context, agents risk overlooking constraints, making costly mistakes, or consuming excessive computing resources as they attempt to infer what they need to know.
To address this, Shah advocates connecting AI to the processes and data enterprises already use. He describes a “contextual layer” that helps agents understand workflows, operate within defined guardrails, and produce recommendations that humans can review and approve. He recommends organizations start with a specific business decision, identify its most difficult steps, and test whether AI can improve them, measuring gains in speed and value before scaling.
Shah also emphasizes the need for workforce education so employees can understand AI outputs and their implications, rather than simply handing decisions to agents. He sees opportunities for context-aware AI to extend from routine operational tasks to enterprise-wide planning in areas such as finance, workforce management, marketing, and sales.
In an interview with MIT Technology Review, Shah frames the challenge as one of “decision intelligence,” where AI augments or automates parts of decisions by understanding the roles and tasks within organizational processes. He cautions that without context, even advanced AI systems may fail to deliver meaningful commercial value or could incur unnecessary costs.
- Nov 10, 2026 · Wired
Kora, a Linux-based desktop OS from Cyanogen cofounder, targets AI power users with local model integration
Trust76 - Nov 10, 2026 · One Useful Thing — Ethan Mollick
Mollick argues new AI models approach "jagged AGI" across coding, business, and humanities
Trust78 - Nov 9, 2026 · The Verge — AI
Former DOJ antitrust chief launches AI legal-research startup to challenge monopolies
Trust72