Only 19% of organizations run multi-agent AI systems today. That number sits at the center of one of the most significant gaps in enterprise technology right now — the distance between what companies say they want and what their operations can actually support.
According to the report, based on a survey of more than 1,600 global business leaders, 85% of enterprises want to become agentic within three years. Yet 76% admit their operations cannot support it. The ambition is nearly universal. The infrastructure is not.
Patrick Thompson, global SVP of customer transformation at Celonis, frames the shift plainly. “Nine in ten leaders are already using or exploring multi-agent systems, so the will is absolutely there, but ambition without infrastructure doesn’t get you very far,” he says.
The Process Gap Blocking AI Returns
The question companies are now asking has changed. With 85% of teams already using generative AI tools for everyday tasks, whether the technology works is largely settled. The harder question — why it isn’t working the way enterprises need it to — points somewhere less comfortable: internal structure.
Siloed teams. Systems that don’t communicate. AI that performs well in a demo and stalls inside a real enterprise environment. The Celonis 2026 Process Optimization Report identifies siloed teams as the top blocker, cited by 54% of respondents, followed by a lack of coordination between departments at 44%. Only 6% of leaders name resistance to change as a hurdle — a figure that almost certainly understates the problem.
The deeper issue is that messy, disconnected processes have long been tolerated because they still produced results — slow, opaque results, but results. Growth masked the dysfunction. AI removed that cover. When 82% of decision-makers say AI will fail to deliver ROI without a proper understanding of how the business runs, sub-optimal processes stop being an operational inconvenience and start blocking strategy entirely.
Context AI Cannot Guess At
For an AI agent to act autonomously and effectively, the report states it needs more than clean data. It needs operational context: how KPIs are defined and calculated, what internal policies govern decisions, where real decision authority sits, and how the organization is actually structured. That knowledge is typically distributed across departments that have built their own systems and languages over years. They don’t share a common operational understanding.
Thompson describes dropping AI into that environment as something like inserting someone into a conversation that has been running for years without giving them any of the backstory.
Process intelligence, the report argues, becomes the connective layer — a shared operational language that grounds AI decisions in how the business actually functions. Without it, agents are, in Thompson’s framing, guessing.
“Organizations that have invested in modernizing their data, systems, and processes are in a far stronger position to enable AI at scale,” he says. “You can’t bolt AI onto a broken process and expect it to work.”
The data reinforces that the problem is not primarily technical. A full 93% of process and operations leaders say process optimization is as much about people and culture as it is about tools. The report positions enterprise AI readiness less as a software question and more as an operating model question — one that requires redesigning how teams, systems, and decisions connect before agents can do anything meaningful with the access they’re given.
Photo by Fernando Narvaez on Pexels
This article is a curated summary based on third-party sources. Source: Read the original article
