Celonis Chief Scientist: Enterprise AI Needs Operational Context to Avoid Costly Mistakes

Why context matters for enterprise AI

According to MIT Technology Review, enterprise AI systems can generate recommendations and automate tasks, but without understanding how a business operates—including its constraints and relationships—they often miss the mark. AI agents acting in isolation may overlook critical workflow dependencies, leading to decisions that are inefficient or even counterproductive.

Devavrat Shah, Chief Scientist of Enterprise AI at Celonis, emphasizes that AI agents need to grasp the role of each task within broader business processes. Without this context, they risk ignoring key constraints, wasting computing resources, or making errors that could have been avoided with a clearer view of operational realities.

Celonis specializes in process intelligence, which digitizes and optimizes enterprise workflows. The company provides AI agents with a real-time "digital twin" of an organization, allowing them to operate within defined guardrails and produce recommendations that teams can review and approve.

How organizations can implement context-aware AI

Shah advises a pragmatic approach to adopting AI in enterprises. Organizations should start with a specific business decision, identify its most challenging steps, and test whether AI can improve them before scaling. This method helps measure actual gains in speed and value, rather than assuming AI will work out of the box.

A critical step is connecting AI to existing processes and data. This ensures agents understand workflows and operate within safe parameters, reducing the risk of errors. Shah notes that governance plays a key role here: defining clear roles for AI agents and certifying them for specific tasks helps maintain control over automated decisions.

Human roles are also set to shift. As AI takes over repetitive, manual tasks, employees may move from execution to more strategic decision-making. However, realizing these productivity gains will require investment in workforce education, ensuring teams can interpret AI outputs and understand their implications.

What happens without operational context

MIT Technology Review illustrates the pitfalls of context-free AI: systems may apply generalized principles to specialized scenarios, failing to account for the nuances of enterprise operations. This can lead to inefficiencies, such as excessive computing resource use for simple tasks, or recommendations that sound plausible but ignore practical constraints.

Shah adds that AI agents without access to specific context are essentially operating in a vacuum. For example, a resource-planning task that should take minutes might instead consume hours of computational effort—or produce results that teams cannot trust or act on.

What this means

Lazyfounder analysis — our interpretation, not reported fact.

For founders and operators, the takeaway is clear: enterprise AI cannot be treated as a plug-and-play solution. The technology’s success depends on grounding it in the realities of how a business actually functions. This means integrating AI with existing workflows, testing its impact on specific decisions, and ensuring teams are equipped to work alongside it.

Companies like Celonis are betting on process intelligence as the missing layer for AI adoption. For startups building AI tools, this signals an opportunity—and a warning. AI that operates in isolation is unlikely to deliver sustained value. Instead, the focus should be on systems that understand and adapt to the unique rhythms of an organization’s operations.

Key takeaways

• Context-aware AI reduces the risk of costly mistakes by grounding decisions in real-world workflows and constraints.
• Start with specific business decisions, test AI’s impact on difficult steps, and measure gains before scaling.
• Integrating AI with process intelligence (e.g., Celonis’ digital twin) helps agents operate within guardrails and produce actionable recommendations.
• Workforce education and governance are critical to ensure teams can interpret AI outputs and maintain control over automated processes.

Related on Lazyfounder

• MIT Technology Review
• More ai news
• Why AlphaGo’s Reasoning Still Outperforms Today’s LLMs—and What’s Missing