AI Agents for ServiceNow: ITSM Now Assist Pricing and Alternatives

ServiceNow reworked its licensing in April 2026, folding the old five ITSM tiers into three AI-native ones, Foundation, Advanced, and Prime, each of which bundles Now Assist and unlimited Virtual Agent into the seat. That sounds like AI is now free. It is not. The AI runs against consumption Assist pools, and different actions cost different amounts: industry breakdowns put a large agentic action at roughly 150 assists and a small one at about 25. Burn through your pooled allowance and you are into per-unit overage, negotiated at renewal.

None of this is published. ServiceNow prices every tier, every pool, and every product through sales, so the numbers you plan around are estimates until a quote lands. Seat licensing alone is commonly cited from about $70 a user at Foundation to $160 to $200 and beyond at Prime, before a single assist is consumed. For a large service desk running heavy deflection, the consumption line is not a rounding error, it is a second budget you renegotiate annually. A flat-rate marketplace agent trades some native depth for a number you can actually forecast.

A marketplace agent does not replace ServiceNow. It plugs into it. You issue a scoped API token, choose which tables, categories, and actions the agent can reach, and it works your tickets the way a service-desk agent would: reading the incident, checking the CMDB and knowledge base, resolving what it can, and routing and tagging what it cannot. Your assignment rules, SLAs, and reporting keep running because the records stay in ServiceNow.

Connecting from outside also buys reach. Now Assist knows what is in your ServiceNow instance. A marketplace agent can additionally read the systems where the real answer lives, your identity provider, your monitoring stack, your HR system, so it resolves the requests a platform-only agent has to escalate. Teams running a busy service desk often pair this with a broader customer service agent for external support, and lean on general enterprise AI agents for the work that spans several systems.

Now Assist is priced and scoped by ServiceNow product line, so what you can automate depends on which modules you license. On IT Service Management (ITSM) the agent works incidents and requests, the password resets, access requests, and how-do-I tickets. On Customer Service Management (CSM) it answers external customer cases from your knowledge base and order data. On HR Service Delivery (HRSD) it handles employee questions about benefits, policy, PTO, and onboarding tasks. Each is a separate ServiceNow product with its own licensing, so a native rollout across all three means three sets of seats and three consumption lines to forecast.

A marketplace agent reaches across those queues through the same scoped API connection. Point it at the ITSM tables and it triages incidents; give it CSM case access and it resolves customer questions; connect it to HRSD and it answers employee requests, often pulling the real answer from your HR system rather than only the knowledge base. Because it connects from outside on a flat subscription, you are not buying and metering Now Assist separately for ITSM, CSM, and HRSD. Many teams run this alongside a dedicated AI agent for HR for people operations and an AI agent for IT support for the help desk.

The fastest way to lose a service desk's trust is an agent that confidently closes the wrong ticket or grants access it should not. So the default is narrow: the agent resolves informational and low-risk requests on its own, and anything that changes production, grants access, or touches a privileged account waits for a person. You open those up deliberately, one category at a time, after you have watched the agent work for a few weeks.

Accuracy is a review discipline, not a vendor promise. Every resolution the agent posts is logged with the knowledge article or CMDB record it drew from, so when a ticket is closed wrongly you can see exactly why and fix the source. Every agent listed on the marketplace is reviewed and security-scanned before it goes live, and you can see exactly what it does and which permissions it needs before you connect it to a production instance. Compare the service-desk agents side by side in the AI agent store, and read how to vet an AI agent before you grant API access.

If you are evaluating ServiceNow's own agent tooling, three things get discussed as though they were one product, and they answer different questions.

• The prebuilt agent library is what you deploy without building anything. ServiceNow ships prebuilt agents across ITSM, CSM, HRSD, and SecOps, and for a standard workflow on a reasonably standard instance they are the fastest path to something running.
• AI Agent Studio is where you build and customize your own, describing what the agent should do in natural language and assembling it from templates and prompt steps. It is genuinely low-code, and it is also where a "quick pilot" turns into a platform project, because a custom agent needs someone to own it, test it against your data model, and maintain it after every upgrade.
• AI Agent Fabric is the connectivity layer. It uses Model Context Protocol (MCP) to let agents reach tools and systems outside ServiceNow, which matters because a large share of service-desk answers live in systems ServiceNow does not own.

The practical decision for most teams is not Studio versus a competitor. It is whether the workflow you want automated is close enough to a prebuilt agent to just turn on. If it is, do that. If it is not, you are choosing between a build in Studio, staffed and maintained by someone on your platform team, and a ready-made agent connected through the API on a flat subscription. The build makes sense when the workflow is genuinely specific to your organization and you have platform capacity. It makes much less sense when the job is ordinary first-line deflection that hundreds of other service desks also run.

The cost side pushes in the same direction. AI features are not part of a base ServiceNow licence: unlocking them means the higher Pro Plus or Enterprise Plus tiers, and the generative work then draws down consumption Assists on top. That combination, a tier upgrade plus a variable meter, is the reason ServiceNow AI budgets are hard to forecast in year one. A connected agent has one published price and no tier prerequisite, which is why it is often the right way to prove the use case before committing to the platform upgrade.