A custom MIT Technology Review Insights report says companies must redesign processes, data access, and controls before AI agents can work across the business.
Enterprise AI can perform useful tasks inside separate departments, yet those systems may still fail to share what the wider business knows. A sales system might not know about an open support ticket, while a marketing system lacks information already held by finance.
The result can be a collection of capable tools rather than one connected operation. The report says the next step is not simply choosing a stronger model. Companies must change how people, processes, and data work together.
The report calls this move from AI as a tool to AI as an operating model the “agentic shift.” An operating model is the way an organization makes decisions and gets work done. For AI agents—software systems designed to carry out multistep tasks—to work across departments, the report says companies need real-time connections, governance, and control.
The report was produced by Insights, MIT Technology Review’s custom-content arm, not its editorial staff. Its recommendations are a planning framework, not independent proof that these practices have already improved business results widely.
Redesign the work before choosing the model
The report’s first recommendation is to redesign business processes before selecting an AI model. In practical terms, a company would begin by examining how work moves between teams, where decisions happen, and which information each step requires.
For example, a company could map how a customer issue moves between sales, support, and finance. It might then decide where an AI system should gather information, suggest a next step, or take an action automatically. This customer-issue workflow is an example of how to apply the report’s idea; the report does not describe or test this specific workflow.
The report says process redesign should come before model selection. It recommends building for how the technology may evolve instead of retrofitting roles and workflows after deployment.
That changes the main planning question. Rather than asking only which model to buy, a company would first ask which process should change and what role AI should play in it.
The report describes “process-first” companies as better positioned to generate sustained returns. However, the report does not establish how widely these practices have been deployed or whether they improve performance across businesses. That makes the claim useful as a proposed operating principle, not a proven formula.
Make data ready, not merely plentiful
The report’s second recommendation concerns data infrastructure. It distinguishes between owning large amounts of data and having data that AI systems can use when a decision is being made.
In practical terms, data readiness can involve making information accessible in the right setting, with suitable permissions, structure, and workflow context. Those details are a practical interpretation of the report’s recommendation, rather than a complete list of requirements stated in it.
The report specifically calls for rebuilding data infrastructure around accessibility rather than volume. The goal is to help approved systems find and use information already spread across an organization, instead of treating a larger central data store as the answer.
It proposes a “sovereign, composable” foundation. In plain English, this means a flexible setup that can query and prepare data where it already resides, without requiring the company to migrate or centralize every record.
That approach matters when information is spread across cloud services, internal systems, or different legal jurisdictions. The report connects this design to data residency laws and says companies need control over where models run, where data lives, and who controls it.
The practical question is therefore not only, “How much data do we have?” It is also, “Can the right system use the right information without breaking company rules or residency requirements?”
Replace fixed technology stacks with flexible parts
The report also recommends replacing fixed technology stacks with composable architectures. A composable architecture uses parts that can be replaced or rearranged, rather than depending on one permanent platform.
That flexibility is intended to help organizations adapt as models and tools change. A company might need to replace a model or data connector without rebuilding every related system.
This does not mean flexibility removes the need for careful design. A system that connects more data and processes also needs controls for how that access works.
The report says the shift requires governance and control so organizations can act on shared intelligence reliably. In practice, a company would need to define which information an AI agent may access, which decisions it may influence, and which actions require human involvement. These are practical ways to apply the report’s governance principle, not documented results from a tested system.
Spending does not solve disconnected systems
The report places its recommendations against a backdrop of rapid investment. It says global AI investment is set to reach $2.5 trillion in 2026, up 44% from the previous year.
More spending and stronger models do not automatically connect departments. If each system remains in its own silo, an organization may add more automated activity without creating a shared view of customers, operations, or decisions.
That is the report’s answer to the reader’s question: redesign the process first, prepare data for access where it resides, and use flexible technology with clear control over actions and jurisdictions.
For an enterprise planning its next AI project, the useful next step is to choose one cross-functional workflow and map its people, decisions, data, and controls. Only then should the company compare models and decide what the AI system may safely do.