Agent sprawl drags enterprise workflow integration down 14 points

That capability matters more than the composite score does, because connecting work across functions is what separates an organisation running AI in several departments from one running it as a business. The report measures AI-enabled workflows at 40, the lowest of its seven pillars and 17 points behind vision and leadership at 57, and explains the year-on-year drop by pointing to organisations that still operate fragmented platforms with a new wave of agent sprawl sitting on top of them. Agents were added to estates that had never been unified, and the additions have made those estates harder to connect than they were before the agents appeared.

The speed of that accumulation explains a good deal. A year ago, 71% of executives told the same survey they were not very familiar with agentic AI, and this year 59% of organisations report using it with a further 30% piloting, which the report describes as among the fastest adoption curves in its three-year history.

Progress on the work those agents were bought to perform has moved at nothing like the same rate, with 9% reporting meaningful progress towards autonomous, multistep workflows, 5% redesigning work instead of automating what already exists, and none of the organisations surveyed having built a cross-functional, self-improving agentic operating system.

“Organisations are moving agents into production faster than they’re building the capabilities to manage agents, both individually and collectively,” said Vijay Kotu, Chief Analytics Officer at ServiceNow, who described the problem as a deployment gap and argued that most enterprises are underestimating how early the architectural decision needs to be taken.

The new layer compounds the debt the old one created

Only 16% of organisations have widely or fully replaced fragmented legacy systems with an integrated platform, which leaves the majority attaching software capable of acting on its own to architectures that cannot give it a complete, real-time view of the business it is acting inside. Early wins in that environment tend to stay where they started, difficult to replicate beyond the team or function that produced them, and the report notes that this is precisely why isolated proof points keep failing to scale into anything enterprise-wide.

What distinguishes this round of integration failure from previous ones is that the new layer multiplies its own errors. “Connected agents amplify each other’s errors as readily as they amplify each other’s value,” Kotu said. “Complexity compounds faster than most organisations anticipate, and by the time it becomes visible, the cost to correct is significant.” He drew a comparison with cloud and API governance, where organisations that invested early paid a one-time architectural cost while those that delayed paid repeatedly at increasing scale, and warned that partial remedies will not hold, since “siloed orchestration is still fragmented execution.”

Employees can see the outcome from inside. Fifty-eight percent say their organisation is not doing a good job of connecting AI-enabled workflows across the enterprise, and 41% rank data silos among their organisation’s biggest AI mistakes, a figure matched almost exactly by the 41% of organisations that cite siloed data as a major barrier to adoption.

Governance is now the constraint on capability already purchased

Twenty percent of organisations have implemented AI testing, auditing and risk assessment processes, which leaves four in five running or piloting agentic systems without the controls that would let those systems be trusted with anything consequential. Inadequate data accuracy, access and management remains the largest reported barrier to adoption at 71%, with concerns over transparency and misinformation cited by 59% of executives and regulatory complexity by 56%.

The research attributes slow movement on autonomous execution to risk avoidance, observing that agents without guardrails can take unintended actions across interconnected workflows faster than anyone can catch them, and that reasoning holds. It also means the governance deficit has become the binding constraint on capability organisations have already paid for, since the 59% using agentic AI cannot safely extend it into the multistep territory that produces the returns.

Holly Briedis, SVP, Global Industries and Solutions at ServiceNow, argued that the sequencing problem starts earlier than governance. “Automating a broken process doesn’t create value. It simply makes the problem happen faster,” she said, adding that the leaders getting this right take a step back before deploying a single agent and ask whether the process reflects the organisation they are building towards or the one they inherited.

The leading group is separated by architecture, and the distance is measurable

Around 21% of organisations qualify as what the report calls Pacesetters, scoring above 60 on the index and averaging 74 against 45 for everyone else, and their defining characteristic is orchestration, connected data and governance installed ahead of deployment. Fifty-eight percent of them have streamlined and integrated workflows across business functions with AI, against 5% of others, and 36% use agentic AI for autonomous multistep workflows, against 2%, which is where the composite gap in maturity scores turns into an operating difference.

The returns track those choices, with Pacesetters reporting an average AI return on investment of 160% today, projected to reach 194% within two years, alongside 5.6 times higher productivity, 2.7 times greater ability to scale and 2.6 times better performance on risk reduction.

“Embedding AI into core workflows resulted in performance gains, rather than introducing standalone tools,” said the CIO of a Pacesetter insurance company in France, describing a decision most organisations in the study have yet to take.

Executives are misreading the willingness of their own workforce

Surveying employees alongside executives for the first time produced the study’s most inverted set of findings, with 71% of employees expecting AI to improve morale and job satisfaction against 45% of executives, and 67% expecting it to enable focus on higher-value work against 46%. On the risks the positions reverse, since 34% of executives worry about skills erosion through AI dependency compared with 18% of employees, and 37% raise moral and ethical concerns compared with 23%.

The report concludes that executives are overestimating cultural resistance while underestimating readiness, and the organisations acting on that misreading are underinvesting in people who have already made up their minds, with 42% of employees saying they are not receiving enough AI training, 59% of organisations holding no long-term HR plans to support the future of work, and 21% having assessed AI skills across the enterprise.

“Leaders should be skilled in guiding this change, assisting employees in viewing AI as a tool for empowerment rather than a threat,” said one UK employee quoted in the research. Bhavin Shah, Senior Vice President and General Manager of Moveworks and AI at ServiceNow, set out what happens when the design ignores them: “When people have to work for the system rather than the system working for them, they don’t use the AI. They work around it. And the moment they work around it, ROI plummets.”

Local pressure points vary while the structural problem does not

Government organisations increased AI spending 140% year on year, more than any other sector surveyed, with manufacturing close behind at 126%, and the obstacles that money meets differ considerably by market. Legacy infrastructure weighs heavier in India because many critical services were digitised late on low-cost stacks now bound tightly to national platforms, while in Japan consensus-driven decision cycles conclude after the technology has moved on, and in France only 42% of executives say clients are happy with AI-enabled experiences against 63% in the US.

The UAE findings released this week reproduce the same pattern at national scale, with AI spending up 105% year on year and maturity climbing 13 points to 48 out of 100, while 14% of organisations have replaced legacy systems with integrated platforms and 7% have used agentic AI to build autonomous workflows. “The organisations pulling ahead are no longer distinguished by how much they spend on AI, but by how effectively they operationalise it,” said Saif Mashat, VP, Middle East and Africa at ServiceNow, who described the shift required as one from point solutions to unified, orchestrated platforms.

The two-year outlook leaves the gap open

Looking two years ahead, 20% of organisations expect to be using agentic AI to create autonomous multistep workflows, which suggests that the majority intend to keep buying capability they have no near-term plan to connect, and that the integration measure which fell this year has no particular reason to recover next year either.

Brian Solis, Head of Global Innovation at ServiceNow, located the unfinished work in organisational design. “Every leader reading this report has an org chart. Very few have a work chart, a living map of how outcomes flow and could flow through the enterprise, showing where humans add irreplaceable imagination and judgment and where agents can own execution, securely, end to end,” he said, adding that these are organisational design questions no enterprise has fully answered.

Amit Zavery, President, Chief Product Officer and Chief Operating Officer at ServiceNow, stated in his introduction to the research that most organisations are still automating yesterday’s work instead of reimagining tomorrow’s, and the year’s numbers suggest a growing number are doing so on infrastructure that grew less connected while they were at it.