Gartner projects that more than 40% of agentic AI projects will be cancelled by 2027, citing cost overruns, inaccuracy, and weak governance. A new survey of 500 senior IT leaders reinforces why: most organizations have not built the operational foundation that AI actually requires to move from pilot to production.
The research, conducted by MIT Technology Review Insights in partnership with Celigo, surveyed IT leaders at mid- to large-size companies across the United States in December 2025. All 500 respondents were actively pursuing AI in some form. The findings point to a consistent pattern: integration infrastructure, not AI capability, separates organizations that succeed at scale from those that stall.
Progress exists, but it is uneven
Three in four surveyed companies (76%) report at least one department running an AI workflow fully in production. That is a meaningful shift from earlier years, when enterprise AI success was largely theoretical. Still, the distribution of that success is narrow.
Nearly half of organizations (43%) find AI works best when applied to well-defined, already-automated processes. A quarter have managed to apply it to new processes. One-third (32%) are spreading AI across various workflows, with mixed results.
The gap between departmental wins and enterprise-wide adoption remains wide. Most organizations are not structured to close it.
Governance and ownership are fragmented
Two-thirds of organizations do not have a dedicated team responsible for maintaining AI workflows. Only 34% have built one. Responsibility is scattered: 21% leave it to central IT, 25% assign it to departmental operations, and 19% spread accountability across multiple groups with no clear owner.
That fragmentation matters more as AI systems become more autonomous. Agentic AI, which executes multi-step tasks with limited human oversight, demands clear accountability structures. Without them, errors compound and governance breaks down quietly before anyone notices.
Integration platforms change the outcome
The strongest predictor of advanced AI implementation in the survey is not the sophistication of the AI model. It is whether the organization has an enterprise-wide integration platform.
Companies with that foundation are five times more likely to draw from diverse data sources in their AI workflows. Six in 10 (59%) of those organizations use five or more data sources. Among organizations using integration only for specific workflows, that figure drops to 11%. Among those with no integration platform at all, it reaches 0%.
The same pattern holds for multi-departmental AI rollout and for organizations’ confidence in granting greater autonomy to AI systems going forward. Integration does not just support AI. It determines how far AI can realistically go inside a business.
The missing layer
The report argues that without connected data, stable automated workflows, and clear governance models, AI initiatives tend to remain trapped in pilots. The technology itself is not the obstacle. The operational infrastructure around it is.
- 76% of surveyed companies have at least one AI workflow in production
- 43% succeed by applying AI to well-defined, existing processes
- Only 34% have a dedicated team for AI workflow maintenance
- Companies with enterprise-wide integration platforms are 5x more likely to use diverse data sources
- 59% of those companies use five or more data sources in AI workflows
As agentic AI expands the autonomy of automated systems, the tolerance for structural gaps shrinks. Organizations that treat integration as a foundational requirement rather than a technical afterthought are the ones consistently reaching production at scale.
Photo by Blake Connally on Unsplash
This article is a curated summary based on third-party sources. Source: Read the original article
