Key Points
- AI agents can only act reliably when ServiceNow runs on clean data, simplified workflows, dependable integrations and clearly defined permissions.
- Modernizing the ServiceNow core requires assessing technical debt, rationalizing workflows and establishing reusable governance before introducing autonomous operations.
- A healthier platform can reduce manual ticket handling by 40%, modernize the core within three to six months and double workflow deployment speed.
I’ve had multiple conversations with senior customer leaders this year where the ask was almost word for word the same: “When do we get AI agents doing real work, not just answering questions?”
It’s a fair question to be asked in 2026, but it’s the wrong place to start. Jumping straight to it, before the groundwork is ready, is exactly why so many AI initiatives quietly stall six months in.
There’s a more basic question sitting underneath all of it, and it’s the one most roadmaps skip: is your ServiceNow platform actually ready for AI? For organizations that have been running ServiceNow for several years, the answer is usually more complicated than the roadmap assumes.
A Number Worth Sitting With
At ServiceNow’s own Knowledge 2026 event, CEO Bill McDermott cited a stat that matched what I’d already been seeing on the ground: despite enterprises collectively pouring billions into AI, ServiceNow’s own Enterprise AI Maturity Index found only around 16% have actually managed to embed AI workflows across their business functions. 2025 was the year everybody got to experiment. Very few made it past that.
The reason isn’t usually the AI itself. Almost every time I dig into a stalled rollout, the real issue is the foundation it was asked to stand on.
Take a customer organization that’s been running ServiceNow for five, six, seven years. Most of these platforms didn’t get complicated on purpose. A team needed a workflow, so a workflow got built. A business unit needed an exception, so a customization got added. Another system needed an integration, so another interface got stitched in. Multiply that by several years and pull back the curtain: you’ll usually find workflows customized so many times nobody remembers why, integrations quietly pointing at systems that got decommissioned two reorgs ago, and a CMDB nobody’s fully trusted since the last major upgrade. None of that is anyone’s fault, it’s just what happens to a platform that’s been genuinely useful for a long time. But it turns into a real problem the moment you hand decision rights to something that acts on its own.
The platform usually still works, tickets get logged, workflows complete, dashboards report their numbers. But underneath, a few signs tend to show up in almost every mature instance:
- New workflows take months rather than weeks to deploy, and developers spend more time maintaining old customizations than building new value.
- Multiple workflows quietly perform the same function, and integrations become difficult to understand, let alone govern.
- Manual triage continues despite prior automation investment, and AI initiatives require far more preparation than anyone budgeted for.
- Platform teams grow reluctant to change anything, for fear of breaking something else.
At that point, the instance has effectively become a digital estate that needs modernizing, not a platform waiting for its next feature.
An agent that resolves an incident or approves a request without a human checking its work is only as good as the data and permissions underneath it. If your instance can’t reliably tell that agent which CI is actually live, which workflow is the real one, or who’s allowed to sign off, you don’t have an AI strategy. You have a fast way to make bad calls at scale.
AI Does Not Eliminate Technical Debt. It Exposes It.
ServiceNow’s own 2026 roadmap is increasingly built around autonomous work: AI agents, governed execution, connected enterprise data, and workflows that can move from recommendation to action. Capabilities like Workflow Data Fabric exist specifically to give autonomous AI better context to work with, but an agent still needs something reliable to act on: clean data, clearly defined processes, real permissions and controls, workflows worth reusing, integrations that actually work, and governance wrapped around all of it.
Put simply: AI readiness is fundamentally a platform-readiness problem, not a shopping decision.
Governance, Not Generation, is Where the Real Conversation is
The center of gravity in almost every one of these customer conversations isn’t “look what our AI can generate” anymore. It’s governance. The customers furthest along describe their AI agents as systems that take real action across workflows, a step beyond conversational assistance and they’re unanimous that governance is what actually gates whether that gets signed off internally. More than one customer CIO has put it to me in almost the same words: the ambition is for ServiceNow to become the operating system their enterprise AI agents run on. That’s a big claim, and it only holds up if the platform underneath can enforce permissions, keep an audit trail, and constrain exactly what each agent is allowed to touch.
“44% of enterprise AI leaders have only moderate confidence that AI agents can act autonomously without human oversight. That hesitation isn’t nerves it’s a rational response to instances that were never built with autonomous execution in mind.”
Ask a Better Question First
Most organizations start by asking, “Where can we deploy an AI agent?” The better question is: “Which process should become autonomous, and is our platform ready to support it?” That change in framing rewrites the sequence and in my experience it happens in four moves.
- Understand the core
Run a technical-debt and configuration health scan: custom code, unused workflows, redundant integrations, duplicate configurations, process bottlenecks, data-quality issues, ownership gaps, and opportunities to move closer to out-of-the-box ServiceNow. The goal isn’t to “clean up ServiceNow”; it’s to understand what’s actually preventing the platform from becoming an automation engine. - Simplify before automating
Not every workflow deserves to be automated. Some should be redesigned, some consolidated, some retired outright, some returned to standard capabilities. Every unnecessary customization removed is less future maintenance, less testing, and less upgrade complexity. - Build the foundation for agentic operations
Once the core is healthier, look for high-volume, rules-driven processes where an agent can take meaningful action: incident triage, password and access requests, knowledge-driven resolution, common L1 requests, employee support, security operations, routine fulfillment. Sequence by clean data and transaction volume, and start where a mistake is small and recoverable. The objective isn’t to add an agent because AI is fashionable, it’s to remove human effort from work that never needed human judgment. - Put governance around autonomy
An agent that recommends an action is fundamentally different from one that executes it. Enterprises need clear controls around what an agent can access, what it can execute, when human approval is required, how decisions get logged, how performance is measured, and how exceptions are handled. These controls should be built once and reused, so every new agent inherits the same permission and audit model instead of each team improvising its own.
Get the order right, and the payoff shows up in quarters, not in a five-year transformation deck:
40%
Reduction in manual ticket handling3-6
Months typical time to a modernized core2x
Faster new workflow deploymentThe Real Business Case Isn’t “More AI”
For CIOs and platform leaders, the business case shouldn’t be “Let’s invest in AI.” It should be “Let’s remove the friction preventing our ServiceNow investment from delivering more automation.” A modernized core means faster deployment, lower maintenance, and more automation. Once that foundation is in place, AI stops being a chatbot bolted onto a process and becomes part of the operating model itself.
Slower on Purpose Beats Fast and Wrong
I know how “clean up the instance first” sounds, like the cautious answer while a competitor races ahead. In practice it’s the opposite. The enterprises I’ve seen skip straight to deploying agents on years of technical debt end up back at square one: pulling agents out, rebuilding trust with the board, and doing the cleanup they should have done first, except now with a credibility problem attached.
2025 rewarded experimentation. 2026 is rewarding the instances that got their core right first and making the ones that skipped that step explain to their board why the pilot never became a program. If you’re being asked to hand over an AI agent roadmap this quarter, the highest-leverage move probably isn’t a new agent. It’s an honest look at what your platform is actually standing on.
The Sutherland Perspective
At Sutherland, we believe the path to autonomous operations starts with the core.
Our approach combines ServiceNow configuration and technical-debt assessment, instance modernization, workflow rationalization, AI readiness and governed AI-agent adoption.
The objective is simple: Turn a complex ServiceNow instance into a cleaner, faster and more automation-ready service engine.
Because the real question enterprises should be asking isn’t whether you’re ready to buy the next AI capability. It’s whether your ServiceNow platform is ready to use it.
Nishant Agarwal
ServiceNow Practice Head
