TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
MIT Technology Review’s Oct. 22, 2026, Business Lab report examines why enterprise AI systems need operational context to work within real business processes. Celonis chief scientist Devavrat Shah argues that companies should start with specific decisions, measure results, and keep people involved in reviewing AI recommendations.
MIT Technology Review published a report on Oct. 22 arguing that enterprise AI needs information about how an organization’s processes work if it is to support decisions usefully. In the accompanying Business Lab discussion, Devavrat Shah, Celonis’ chief scientist of enterprise AI, said AI agents can assist with tasks inside larger decision processes, but need to understand the constraints and purpose surrounding those tasks.
Shah’s argument is that business decisions are often reached through a sequence of tasks carried out by people and systems. AI may be able to perform or support individual steps, he said, but acting without knowledge of the broader workflow can cause an agent to miss relevant requirements or make recommendations that do not fit the situation. “You can’t just do that in a vacuum,” Shah said in the interview.
The report describes connecting AI to existing enterprise data and processes as one way to provide that context. A contextual layer could help agents identify workflows and operate within defined guardrails, while leaving recommendations open for people to review and approve. Shah also cautioned that agents without adequate context could make costly mistakes or use excessive computing resources while trying to determine what information they need.
For companies deciding where to begin, Shah recommended selecting a specific business decision, finding its most difficult steps, and testing whether AI can improve them. He said organizations can then measure changes in speed and value before deciding whether to expand an application. The report does not provide measured results from a particular deployment or independently validate the approach.
Context Shapes Enterprise AI Decisions
The issue matters as companies look for evidence that AI pilots can support everyday work. A system that produces a plausible answer may still be unhelpful if it does not account for a company’s workflow, responsibilities, or operating constraints. Operational context could affect whether AI recommendations are relevant and whether agents can safely take on parts of a process.
The approach also has implications for employees. Shah said automation and coordination could reduce repetitive manual work and give workers more time for complex decisions. That outcome is not automatic: the report says it would require investment in workforce education, so employees can interpret AI outputs and understand their consequences rather than simply transferring decisions to agents.
For business leaders, Shah’s proposed test—start with a defined decision and measure speed and value—offers a way to evaluate an application before scaling it. The report presents this as guidance, not proof that contextual AI will produce gains across organizations or types of work.
enterprise workflow management software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From AI Pilots to Daily Operations
The report frames enterprise AI as moving from experiments and pilots toward broader use, while noting that leadership teams want clearer evidence of business impact. It focuses on AI agents as tools that may perform individual tasks within existing decision processes, rather than treating each agent as an independent decision-maker.
Shah connects that task-level role to what he calls “decision intelligence”: using AI to support, augment, or sometimes automate parts of a decision process. The report also describes possible future uses across finance, workforce management, marketing, and sales, including resource planning. These are areas Shah sees as potential opportunities, not documented results from deployments presented in the report.
The Business Lab episode was produced in partnership with Celonis, where Shah is chief scientist of enterprise AI. That relationship is relevant when weighing the discussion: the report conveys Shah’s perspective and proposed approach, rather than an independent evaluation of a product or comparative study.
“You can’t just do that in a vacuum.”
— Devavrat Shah, chief scientist of enterprise AI at Celonis
Evidence and Deployment Results
The report does not identify a specific company deployment with measured improvements, nor does it quantify how much context is needed for an agent to make a reliable recommendation. It also does not establish that adding a contextual layer will prevent errors or reduce computing use in every setting. Those outcomes remain claims or possibilities in the material provided, rather than demonstrated findings.
It is also unclear how organizations should compare different ways of connecting AI to operational data, how they should govern access to that information, or what safeguards are needed when an agent acts rather than recommends. Shah’s comments outline a general method—test a defined decision and assess speed and value—but do not specify universal measures or thresholds for success.
Testing Use Cases Before Scaling
The next step proposed in the report is for organizations to identify a concrete decision, isolate the steps that create the most difficulty, and test whether AI can help. Companies would then need to measure speed and business value and decide whether results justify wider use. The report gives no schedule for such tests or a named next milestone.
If companies pursue this approach, the discussion points to two parallel needs: connect AI to relevant processes and data, and prepare employees to assess its outputs. Whether that combination leads to quicker or better decisions will depend on deployment evidence that the report does not yet supply.
Key Questions
What does context-aware AI mean in this report?
It refers to AI systems that can use information about business processes, workflows, and operating conditions when supporting tasks or recommendations.
Why does Devavrat Shah say AI agents need business context?
Shah says a task is part of a larger decision process. Without that surrounding information, an agent may miss constraints or produce an unsuitable recommendation.
What approach does Shah recommend for companies?
He recommends starting with a specific business decision, identifying its difficult steps, testing whether AI helps, and measuring speed and value before considering wider use.
Does the report prove that context-aware AI improves business results?
No. The report presents Shah’s argument and proposed method, but does not provide measured results from a named deployment or independent evidence that the approach works across organizations.
What role do employees have in this approach?
The report says employees may need training to understand AI outputs and their implications, review recommendations, and retain an appropriate role in consequential decisions.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
