How to Think About Agentic AI’s Challenges and Opportunities

Gartner predicted in June 2025 that 40 percent of agentic AI projects would be cancelled by 2027 due to a lack of sufficient controls, resources or clear objectives. In the same report, however, Gartner also predicted that 15 percent of daily work decisions would be made autonomously with the help of agentic AI by 2028. 

What do enterprises do to position themselves to take advantage of agentic AI’s promise?  Executives get excited about the prospects of agentic AI when they hear about its numerous business transformation possibilities. But then they worry about agent-based malware or other new classes of vulnerabilities and attack vectors that are emerging. 

What’s less evident but equally concerning is context scarcity. Despite its name and its popularity with applications builders, Model Context Protocol doesn’t really do more than connect agentic users and models to data resources. It’s a clever wiring protocol, but if the quality, contextualized data agents need to consume for their purposes, doesn’t exist, the agents will fail.

Organizations will be quick to start using MCP, and they will worry less about wiring up and prototyping agentic AI applications, but the problems with data quality and effective, scalable integration and interoperation remain unsolved.