ServiceNow vs. the Agents: Defend or Disrupt?

Table of Contents

• Why ServiceNow Is the Sharpest Test of the Agent Thesis
• The Defense: What ServiceNow Has That Agents Can't Easily Copy
• The Disruption Case: Where the Platform Is Genuinely Exposed
• Now Assist and the Self-Disruption Bet
• The Pricing Knot: Seats, Workflows, and Outcomes
• System of Record vs. System of Action
• What This Means for Buyers, Builders, and the Broader GaaS Market
• Insights Most People Overlook
• References

Why ServiceNow Is the Sharpest Test of the Agent Thesis

Most "is SaaS dead?" arguments get hand-wavy fast. ServiceNow forces precision because it is two things at once. It is a system of record, the authoritative database of incidents, requests, assets, and approvals running inside large enterprises. And it is a workflow engine, the thing that moves a ticket from "opened" to "resolved" across IT, HR, customer service, and increasingly the whole back office.

Agentic AI sold as a service, the broader category this cluster tracks, attacks the second job directly. An agent that can read a ticket, query the relevant systems, take the corrective action, and close the loop is doing exactly what a ServiceNow workflow plus a human agent used to do together. That overlap is why ServiceNow shows up in nearly every disruption conversation about enterprise software. It is also why the company has spent the last two years repositioning itself, in CEO Bill McDermott's words, as the "AI platform for business transformation" rather than a ticketing tool.

Here's the tension in one sentence. ServiceNow makes most of its money from seats, named users who log in to do work. Agents reduce the number of humans who need to log in. If the agent does the work the seated human used to do, the platform's own success metric starts cannibalizing its revenue model. That's not a hypothetical; it's the same seat-erosion dynamic that has SaaS investors nervously watching net revenue retention across the sector.

The Defense: What ServiceNow Has That Agents Can't Easily Copy

Strip away the marketing and ServiceNow's defensive position rests on three things that are genuinely hard to replicate.

The workflow data moat. ServiceNow holds years of structured records about how a specific enterprise actually runs, its approval chains, its escalation paths, its asset relationships, its CMDB. A generic agent doesn't have this. It can be granted access, but the data lives inside ServiceNow's schema, governed by ServiceNow's permissions. This is the incumbent advantage that's hard for agents to cross: proprietary, contextual, permissioned operational data. An agent without that context is improvising; an agent with it is only useful through an integration ServiceNow controls.

Mission-critical embeddedness. When ServiceNow runs your major incident process, ripping it out isn't a software swap, it's organizational surgery. The switching cost isn't the license; it's the retraining, the re-integration with forty downstream systems, and the risk of breaking a process that pages people at 3 a.m. Agents that want to displace this have to be not just better but trustworthy enough to own a process where failure is visible and expensive. As Gartner has noted in its research on agentic AI adoption, trust and oversight remain the gating factors for autonomous enterprise deployment, not raw capability.

Distribution and the trusted-vendor seat at the table. ServiceNow already has the procurement relationship, the security review on file, the master agreement signed. A startup agent has to win all of that from scratch against an incumbent who can simply say "we do that now too." In enterprise software, distribution routinely beats raw capability, the vendor already inside the building gets the first call.

These moats are real. But notice what they protect: the platform's position as the place where work is recorded and orchestrated. None of them guarantee the seat-based revenue model survives intact.

The Disruption Case: Where the Platform Is Genuinely Exposed

Now the other side, because a one-sided take here would be useless.

The clearest exposure is the agent-fulfillment layer. ServiceNow's classic value proposition included routing work to human agents and giving them a console to resolve it. If outcomes get resolved autonomously, the value of that human-facing console drops, and with it the justification for tens of thousands of fulfiller seats. The platform can still orchestrate, but orchestration alone is a thinner, more commoditizable slice than orchestration plus a captive seated workforce.

The second exposure is the browser-and-API agent that routes around the UI entirely. A capable agent doesn't need ServiceNow's interface; it needs ServiceNow's data and the ability to write actions back. If a customer's preferred agent platform, or an internally built one, treats ServiceNow as just one API among many, ServiceNow risks becoming infrastructure: necessary but invisible, and priced like a database rather than an application. That's the "agents as the new UI layer over old software" pattern playing out, and it's quietly one of the more dangerous outcomes for any application vendor, because it strips away the brand and the seat at once.

The third exposure is vertical and horizontal encroachment from purpose-built agents. A vertical agent built specifically for, say, IT incident triage or HR case management can go deep in ways a general platform's add-on may not match at launch. The unbundling risk is real: agents picking off the highest-value, highest-volume workflows one at a time, leaving ServiceNow with the long tail of low-frequency processes that are less profitable to host.

None of these kills ServiceNow. All of them compress the part of its business tied to human seats.

Now Assist and the Self-Disruption Bet

ServiceNow's answer is Now Assist and its broader agentic push, embedding generative and agentic AI directly into the platform so the agents that resolve work are ServiceNow's agents, running on ServiceNow's data, billed by ServiceNow.

This is the incumbent's dilemma in its purest form: cannibalize your own seats before a startup does it for you. McDermott's team has been explicit that they would rather sell you the AI that reduces your seat count than watch a competitor sell it. ServiceNow has paired this with a pricing experiment around AI value, moving toward charging for AI-driven outcomes and usage rather than only per-seat licenses. The strategic logic, echoed across the agent era and detailed in analyses like McKinsey's work on the economic potential of generative AI, is that the value pool is migrating from "software that helps people work" toward "software that does the work." Capturing that pool means pricing the work, not the worker.

The bet is credible precisely because ServiceNow controls the data and the process. An agent running inside the platform inherits the moat. But self-disruption is brutal on the income statement in the short run, and ServiceNow has to manage Wall Street through a transition where the per-seat line might soften before the per-outcome line gets big enough to compensate. That's the tightrope every incumbent repricing for the agent era has to walk.

The Pricing Knot: Seats, Workflows, and Outcomes

This is where the whole debate concentrates, so it's worth being concrete.

Seat-based pricing assumes a roughly stable ratio of humans to work. Agents break that assumption. When one agent can clear the queue that used to need ten fulfillers, the customer's natural move is to cut seats, and ServiceNow's natural defense is to capture the value somewhere else: per resolution, per workflow executed, per outcome delivered, or via a consumption meter on AI actions.

The hard part is measurement and attribution. Outcome-based pricing sounds clean until you ask: who gets credit when an incident auto-resolves, the agent, the underlying workflow, the integration, or the human who designed the playbook? Procurement teams buying outcomes instead of software face a genuinely different evaluation problem, and vendors face a genuinely different forecasting problem. ServiceNow's advantage is that, as the system of record, it is uniquely positioned to measure the outcome, it knows when the ticket actually closed and whether it reopened. Measurement authority may turn out to be as valuable as the data moat itself, because the party that defines "outcome achieved" controls the invoice.

For buyers, the practical takeaway is to model the transition, not the steady state. A move from seats to consumption can lower cost when volume is flat and raise it sharply when volume spikes, and agents tend to increase the volume of actions taken even as they reduce the humans taking them. Negotiate caps and floors now, while you still have leverage.

System of Record vs. System of Action

Underneath all of this sits the structural battle that will decide ServiceNow's fate: system of record versus system of action.

The system of record is the trusted database, the authoritative state of the enterprise. The system of action is whatever actually executes change against that state. Historically ServiceNow was both. The agent era pries them apart. It's entirely possible to have ServiceNow remain the system of record while a different agent layer becomes the dominant system of action, calling into ServiceNow to read and write but owning the customer's attention, workflow logic, and, critically, the budget line.

If that split happens, ServiceNow keeps the data and loses the relationship. It becomes plumbing. The company's entire strategy can be read as an effort to make sure the system of action stays inside its own walls, which is exactly why Now Assist, native agents, and outcome pricing matter so much. They're not features; they're a claim on the action layer.

Whoever owns the system of action owns the customer relationship and the pricing power. That's the prize, and it's why this is a defend-and-disrupt story rather than a pure-defense one. ServiceNow has to disrupt its own seat model to defend its claim on the action layer.

What This Means for Buyers, Builders, and the Broader GaaS Market

For enterprise buyers, ServiceNow is a reasonable place to consolidate agentic work if you negotiate the pricing transition deliberately, insist on transparency about how AI consumption is metered and what happens to your costs as agent-driven volume grows. Don't sign a multi-year seat deal that ignores the agent question; and don't assume "AI-native" challengers can yet match the platform's depth on regulated, mission-critical processes.

For builders of agents-as-a-service, ServiceNow is both a partner and a wall. The opportunity is to be the system of action on top of ServiceNow's system of record. The danger is that ServiceNow closes that gap with its own native agents and treats your integration as a commodity it can replicate. Build where the incumbent's data moat doesn't reach, novel cross-system workflows, niche verticals, judgment-heavy tasks, rather than head-on where ServiceNow already owns the records.

For the broader GaaS market, ServiceNow is the bellwether. If a deeply entrenched, data-rich incumbent can successfully reprice from seats to outcomes and keep its claim on the action layer, the "SaaS is dead" thesis is overstated and the future is incumbents-plus-agents. If it stumbles, if buyers learn to route around its seats faster than it can monetize outcomes, then every seat-based vendor with a shallower moat should be very, very worried.

Insights Most People Overlook

• The measurement authority is worth more than the data moat. Everyone fixates on ServiceNow's CMDB and workflow data. The under-discussed asset is that ServiceNow knows when an outcome actually happened, when the ticket closed and stayed closed. In an outcome-priced world, the party that authoritatively defines "done" controls the invoice. That's a quieter, stickier moat than the data itself, and it's nearly impossible for an external agent to claim.

• ServiceNow's biggest risk is internal, not competitive. The startup-eats-incumbent narrative is overblown here. The real danger is ServiceNow's own sales motion and comp plans, which were built to sell seats. Repricing to outcomes requires retraining a massive go-to-market organization whose incentives point the other way. Most incumbents that fail the agent transition fail at the comp-plan layer, not the product layer.

• Agents may grow ServiceNow's footprint before they shrink it. Counterintuitively, autonomous resolution makes it economical to put ServiceNow workflows in places that were never worth staffing, long-tail processes, smaller departments, mid-market accounts. The seat model couldn't reach those; an outcome model can. The seat erosion story might be partly offset by a workflow-expansion story nobody's pricing in.

• The "thin wrapper" panic cuts the other way for ServiceNow. Pundits worry that agent startups are thin wrappers over foundation models. But an agent that wraps ServiceNow's data and workflow engine is wrapping ServiceNow, not just a model, which means the wrapper's value accrues partly to ServiceNow. The platform can tolerate a thriving wrapper ecosystem as long as it owns the substrate underneath.

• Whoever wins the action layer will be decided by trust, not capability. Capability gaps close fast in this market. What won't close fast is the willingness to let an autonomous agent own a process that pages an executive when it breaks. ServiceNow's two decades of being the trusted owner of those processes is a moat measured in years of earned reliability, and that's the hardest thing for any disruptor to buy.

References

• Gartner, Intelligent Agents in AI and the path to autonomous enterprise deployment
• McKinsey, The economic potential of generative AI: The next productivity frontier