AI
AI agents will take shortcuts. The solution is to design better paths
In 10 November 2026
What happens when an AI agent finds a faster way to achieve its objective, but not the one you intended?
That’s the governance challenge Bruno Billy explores in his Forbes article, “AI Agents Will Take Shortcuts. The Solution Is To Design Better Paths.”
Rather than simply adding more controls, he looks at how organizations can make governed data and processes the easiest path for AI agents to follow.
AI Agents Will Take Shortcuts. The Solution Is To Design Better Paths
As AI agents become more capable of acting autonomously, organizations face a new governance challenge: an agent may accomplish its objective through a path its designers never intended.
Recent tests with frontier AI agents have highlighted this behavior. When agents are pushed to work quickly, efficiently or with fewer tool calls, they can identify shortcuts rather than follow the expected process. Without explicit constraints, organizations cannot assume that an AI agent will naturally choose authoritative data sources, established processes or the appropriate way to handle exceptions.
Governance Creates Friction
Traditional governance relies heavily on policies, controls and monitoring. These mechanisms remain essential, particularly when sensitive data or high-impact actions are involved. But controls also create friction, and friction can influence which path users, and increasingly AI agents, choose.
The challenge is similar to a “desire path”: the informal path people create when the designed route is less convenient than an alternative. Within an organization, governed data may require users to search a catalog, request access and wait for approval, while an uncontrolled spreadsheet may be immediately available. When speed matters, the easier route can quickly become the preferred one.
From the Intended Journey to the Actual Journey
For data governance teams, this creates an important question: why is trusted data sometimes harder to use than an uncontrolled alternative?
With AI agents, the issue becomes even more important. Unlike an experienced employee who may understand the limitations of an unofficial data source, an agent can simply select the route that allows it to accomplish its objective most efficiently.
Governance therefore needs to go beyond preventing agents from taking the wrong action. Organizations should also consider how to make the right action the easiest one to take.
Make the Governed Path Easy
Effective AI governance is not only about governing the agent. It is also about designing the environment in which the agent operates.
Organizations should make authoritative master and reference data easy to discover and consume, trusted data products easier to access than uncontrolled alternatives, and approved business definitions and rules understandable by machines. Agents also need clear routes for common actions and escalation paths when they encounter ambiguity or exceptions.
This approach can also contribute to greater consistency across the organization. When teams and AI agents rely on the same governed data products and business definitions, analyses can begin from a common understanding of core entities such as customers, products and revenue.
Make Friction Intentional
Not all friction should disappear.
The appropriate level of control should reflect the consequences of an action. Retrieving an approved product hierarchy, for example, may require very little friction, while changing banking information, accessing sensitive information or publishing externally may require stronger controls or human approval.
The goal is therefore not simply to add more governance. It is to design better paths.
By making trusted, governed data the easiest and most natural option for both people and AI agents, organizations can build an environment where governance supports action rather than getting in its way.
Build better paths for AI with trusted data
AI agents need more than rules to operate reliably. They need trusted data, clear business context and governed paths that make the right action the easiest action.