According to a new Gartner forecast, approximately 40% of organizations plan to demote or completely abandon their use of AI agents. The primary reason is a lack of effective governance mechanisms, which calls into question the current wave of autonomous system implementation in the corporate sector.
What Happened
Gartner analysts state that a significant portion of current AI agent deployments will face degradation or complete termination. Companies are discovering that existing control approaches either overly restrict the agents' capabilities or leave them without proper oversight, creating unacceptable risks.
Context
The problem lies in the so-called "governance binary trap": organizations must choose between excessive restrictions that render agents useless and full autonomy that leads to uncontrolled actions. To ensure reliability, a transition to multi-level autonomy models is required, incorporating automatic "circuit breakers" and rapid action rollback systems.
Why It Matters for the Industry
Developers of major enterprise software, such as SAP, Oracle, and Salesforce, are being forced to shift their focus from creating simple "agentic" workflows to building complex systems for monitoring, risk management, and legal accountability. The industry expects growing demand for middleware solutions and orchestrators capable of providing controlled autonomy and state management.
Why It Matters for Users
For businesses, this means the current hype surrounding autonomous agents is hitting an operational reliability barrier. In the near future, users and executives will see a transition from simple LLM wrappers to systems with predictable behavior, where the level of autonomy is strictly tied to the security architecture.
What Remains Unknown / Limitations
Expert diagnoses are unanimous, though emphases vary: ranging from the need for purely engineering solutions to increase reliability to strategic changes in enterprise software architecture and the formation of new standards for agent governance.
Sources
Author
Look at AI, Editorial Team