For years, enterprises were encouraged to buy the best application for every job. That saw HR, finance, and customer service all choose their own preferred platforms and tools, with IT teams layering new SaaS products wherever gaps appeared. Employees just learned to put up with disconnected systems, workarounds, and manual processes.
But now AI agents are landing in the middle of those fragmented estates – and according to a growing number of enterprise technology leaders, they are exposing the “AI chaos” that organizations have been living with for years.
At ServiceNow Knowledge 2026 earlier this month, CEO Bill McDermott described organizations running “367 different applications on average, AI bolted onto every one of them like a sidecar”, warning that many environments remain disconnected, ungoverned, and difficult to control.
The concern, therefore, is no longer simply application sprawl. It’s what happens when autonomous systems begin acting across that sprawl.
“Most organizations have more AI in production than they’ve inventoried or accounted for,” said Nenshad Bardoliwalla, group VP of AI products at ServiceNow. “AI agents are taking real actions, moving real money and affecting real people. Most of those agents are running without a system that governs them. That’s not agentic business, that’s agentic chaos.”
The message is significant – AI is no longer being discussed purely as a productivity tool. Increasingly, it’s being treated as something that exposes weaknesses in enterprise architecture, governance, and operational visibility.
From Keynote Messaging to Enterprise Strategy
The shift is already shaping how large organizations structure AI deployments.
This month, BT Group announced a multi-year AI-Ops programme with Accenture built around ServiceNow’s platform, using AI agents to autonomously analyze and remediate operational issues across managed services and cybersecurity environments.
BT said the initiative would allow agents to “self-heal at machine speed while dynamically learning”, with AI-driven operations supporting resilience, security, and automation across customer environments.
At almost the same time, ServiceNow Elite Partner Synechron published research arguing that AI disruption will strengthen deeply embedded enterprise platforms rather than weaken them.
The company described ServiceNow as “the orchestration layer for the entire software stack”, arguing that AI agents require governed workflows, audit trails, and operational context to function safely inside regulated enterprises.
App Sprawl Becomes Data Sprawl
New research from The Futurum Group found 41% of organizations are planning application consolidation initiatives as AI adoption accelerates, while reliance on best-of-breed procurement strategies continues to fall.
The report argues that fragmented application environments create an “integration tax” that makes enterprise-wide AI deployment increasingly difficult.
Chris Gabriel, executive head of marketing and innovation UK&I at NTT DATA Business Solutions, believes many enterprises are now confronting the consequences of years of fragmented software purchasing.
“AI doesn’t replace holes in your application, process and data landscapes, it exposes them,” he said.
Gabriel argues the real challenge is not necessarily application sprawl itself, but what lies beneath it.
“Sprawl of apps equals sprawl of data, different quality of data, different versions of truth,” he said. “Can an AI make a decision confidently or not? If we have functions with different versions of the data truth, how do companies make end-to-end realizable decisions?”
That becomes significantly harder once AI agents begin operating across multiple systems simultaneously.
“Agents just expose that challenge by a factor of X,” explained Gabriel.
Gabriel also said that humans have historically compensated for poor system design in ways AI cannot.
“Humans tend to work around them. AI exposes things whereas humans tend to work around them,” he explained.
From Automation to Autonomy
It’s important, too, to draw a distinction between traditional automation and agentic AI. For years, enterprise automation largely followed predictable paths. AI agents introduce something very different.
“The simple answer is automation did what it was told. AI has the potential to do what it ‘thinks’,” said Gabriel. “Automation follows a rigid path. AI’s power is its ability to work within a window of reasoning.”
Snow Tempest, research manager for IDC’s ITSM Program, said concerns around “agent sprawl” are beginning to emerge as organizations move from experimentation into operational deployment.
“Traditional automation tools were more structured and rigid. AI agents can execute on decisions in pursuit of objectives, which is both a benefit and a risk,” she said.
That risk featured heavily throughout ServiceNow’s keynote demonstrations at Knowledge 2026. Presentations showed AI agents being halted mid-process after prompt injection attacks were detected, with “kill switches” automatically removing permissions before wider damage spread across connected systems.
The focus on runtime governance, observability, and AI oversight could be a sign of how quickly the market conversation is shifting from AI experimentation toward operational control.
Why Orchestration Suddenly Matters
The implication running through much of the discussion is that AI may actually strengthen the position of large enterprise platforms rather than weaken them.
For years, parts of the software industry assumed AI-native tools would reduce reliance on deeply embedded enterprise systems. Instead, many organizations are discovering that AI agents require trusted operational data, workflow visibility, dependency mapping, and governance layers to function safely.
Hardit Singh, technical architect at EY India, said enterprises are increasingly shifting attention away from individual applications and toward orchestration.
“Companies are now thinking less about ‘which app is best?’ and more about ‘how do we orchestrate AI safely across the business?’” he said.
Singh believes platforms such as ServiceNow are becoming more strategic precisely because AI agents require operational context to act responsibly.
“An AI agent can’t just restart a server without knowing what business service depends on it. That’s where CMDB and CSDM become critical.”
Synechron’s report makes a similar argument. Rather than replacing enterprise platforms, the company argues AI may compress value into fewer, deeper systems capable of providing governance, compliance visibility, and operational context at scale.
Gabriel believes the future may lie in combining trusted systems of record with orchestration platforms capable of governing AI-driven activity across the business.
“We like the model of A2A. SAP agents are the agents of truth inside finance, but ServiceNow agents become the agents of action taking that truth across the business,” he said.
“I 100% trust my SAP finance system to give me answers about my business. If I can layer that system of truth with a system of orchestration and action and governance, my confidence level in AI is higher.”
Final Thoughts
AI agents are forcing enterprises to confront questions they may have been putting off asking: which systems contain trusted data, which workflows govern the business, who controls AI actions, and how much fragmentation can those organizations tolerate once autonomous systems begin operating at scale?
Whether they ultimately consolidate applications or just govern them more effectively is unclear for now. What does appear increasingly likely, however, is that AI will make it much harder to ignore the cracks running through enterprise technology estates.