Bridging the operational AI gap

A new MIT Technology Review Insights report, sponsored by Celigo and based on a December 2025 survey of 500 senior IT leaders at mid-to-large US companies, reveals a critical inflection point in enterprise AI adoption. While transformational potential is widely acknowledged and 76% of organizations now have at least one department with an AI workflow fully in production, the path to scalable, enterprise-wide success remains fraught. The research highlights a growing 'operational AI gap' where pilot projects struggle to transition into stable production systems. Gartner's stark prediction that over 40% of agentic AI projects will be cancelled by 2027 due to cost, inaccuracy, and governance challenges underscores that the core issue is not the AI models themselves, but the missing operational and integration foundation required to support them.

Technical implementation data shows that AI succeeds most frequently with well-defined, established processes, with 43% of organizations finding success there versus only 25% with new processes. Crucially, the report identifies enterprise-wide integration platforms as the key differentiator for robust AI implementation. Companies utilizing these platforms are five times more likely to incorporate five or more diverse data sources into their AI workflows (59% vs. 11% for point solutions and 0% for no platform). These organizations also demonstrate more multi-departmental AI deployment, greater workflow autonomy, and more confidence in assigning future autonomy to AI agents. The findings suggest that without this holistic approach to integrating data, applications, and systems—and dedicated teams for maintenance, which only 34% of companies have—AI initiatives risk remaining siloed experiments rather than driving transformational business outcomes.