ServiceNow

Best for: Building autonomous agents that reason over Workflow Data Fabric and CMDB context, plan multi-step actions, and execute via existing Now Platform integrations and Flows. ServiceNow ships thousands of pre-configured agents (incident triage, password reset, onboarding tasks, expense reconciliation, supplier risk) and positions AI Agents as the path "beyond sidecar AI" — agents that *complete* work, not just assist with it.

• Pre-configured agents shipped across ITSM, HRSD, CSM, Field Service, Finance, Legal, Procurement, and industry verticals (telecom, banking, insurance, public sector)
• Plan / act / reflect orchestration grounded in Workflow Data Fabric and Context Engine
• Invokable from EmployeeWorks chat, embedded inside any record page, triggered by Flows, or run on schedules
• Bring-your-own-LLM on Prime tier; Now LLMs and partner models on lower tiers
• Build Agent skills (new) let developers author and deploy from any IDE
• Industry agent libraries for card disputes, insurance claims, telecom service activation

Limitations: Agent quality is heavily dependent on Workflow Data Fabric grounding — without a clean data layer, agents hallucinate or refuse to answer. Pre-built agents skew toward English-language US/EU process patterns; localization and regional process variation often require Skill Kit work. Token consumption per agent run is not always predictable, especially for multi-step plans.

Best for: The authoring surface for custom AI Agents. Anyone — admin, business analyst, citizen developer — can configure a new agent using natural language: describe the goal, attach the data sources, define the actions, deploy. The studio handles plan-loop orchestration, monitoring, and lifecycle management.

• Natural-language agent definition: "build me an agent that answers HR policy questions and files PTO requests"
• Visual canvas for topics, actions, and integrations into Now Platform Flows
• Onboarding wizard with guided instructions; live preview against tenant data
• Built-in monitoring: per-agent run logs, token consumption, success rate, escalation paths
• Reusable skill libraries shareable across an org or via AgentExchange-equivalent partner sharing
• Tight integration with AI Control Tower for governance handoff at deploy time

Limitations: 'Anyone can build an agent' is true; 'anyone can build a *good* agent' still requires understanding of process boundaries, escalation logic, and Workflow Data Fabric grounding. Studio output quality is tied to how well the underlying Now Platform data is modeled. Custom agents can shadow OOTB agents — orgs without naming conventions end up with duplicate or conflicting agent inventories.

Best for: The 'single pane of glass' for every AI agent running in the enterprise — including non-ServiceNow agents reporting in via API. AI Control Tower handles inventory, policy enforcement, observability, cost attribution, and human-in-the-loop escalation. ServiceNow's pitch is that even if you run Agentforce, Copilot, or homegrown agents, you should run AI Control Tower as the governance plane on top.

• Central inventory of all AI agents (ServiceNow + non-ServiceNow via API registration)
• Policy enforcement: which agents can act on which records, which require human approval, which must escalate
• Observability: per-agent run logs, token cost attribution, accuracy and confidence metrics
• Risk classification: dollar-impact tiers, regulated-data tiers, automatic governance gates
• Integration with EmployeeWorks for human handoff when agents hit policy boundaries
• Reporting dashboards aligned to executive AI ROI and risk metrics

Limitations: Sells better when ServiceNow is already the enterprise system of record — orgs without ServiceNow as their IT service backbone get less out of registering external agents here. 'Govern non-ServiceNow agents' capability is real but limited to what the third-party API exposes; closed ecosystems (e.g., Microsoft 365 Copilot) surface only partial telemetry.

Best for: Connecting enterprise data from any source — warehouses, lakes, SaaS apps, file stores — without moving it into ServiceNow. Workflow Data Fabric is the grounding layer that makes Now Assist and AI Agents accurate: when an agent answers a CMDB question, it can also reach into Snowflake, BigQuery, Databricks, or an S3 lake for context.

• Federated query against external warehouses (Snowflake, Databricks, BigQuery, Redshift) without data replication
• Native CSDM (Common Services Data Model) extensions for AI grounding
• Live agent + AI agent unification in a single system of record
• Connectors to CRM, ERP, HCM, observability, and security stacks
• Works as the data spine for Context Engine, Now Assist, and AI Agents
• Tenant-scoped access controls — fabric inherits source-system permissions

Limitations: Federated query is powerful but introduces latency and dependency on source systems being healthy — outages in your Snowflake instance now affect Now Assist response quality. Some connector capability is 'available via partner integration' rather than native, which adds licensing and SLA complexity. Customers expecting it to fully replace existing CDP / data lake investments are usually disappointed; it complements those layers rather than supplanting them.

Best for: The grounding layer that connects relationships, policy, and decision history — so an AI agent doesn't just have facts, it has the *organizational* context that explains why those facts matter. Announced in 2026 as the missing piece between Workflow Data Fabric (raw data) and AI Agents (action). Context Engine answers questions like 'what policy applies here,' 'who approved this last time,' and 'what's the precedent.'

• Graph-based relationships across people, teams, processes, records, policies, and decisions
• Policy-aware retrieval — agents pull the rules that apply to the specific record / region / role
• Decision history and precedent — what was approved before, by whom, with what outcome
• Identity and entitlement context — what the requesting user is allowed to ask for, see, change
• Natural-language explanation surfaces — agents can show their reasoning trace
• Continuous learning from approved / rejected agent actions feeds future grounding

Limitations: Newest of the products listed — capability claims outpace mature deployments as of mid-2026. Org policy modeling is a real consulting effort; out-of-box context coverage is shallow until customer-specific policies are loaded. Heavy dependency on Workflow Data Fabric quality — Context Engine can only graph what the fabric exposes.

Best for: The unified employee-facing chat surface — one place an employee can ask 'where's my laptop request,' 'how much PTO do I have,' 'what's the status of my expense report' — without knowing which department owns which workflow. EmployeeWorks is the human-facing chat side of the AI Agent platform; agents do the work behind it. EmployeeWorks combines the Moveworks conversational layer (ServiceNow acquired Moveworks in December 2025) with ServiceNow's portal and workflows; it went GA on Feb 26, 2026.

• Single chat UI replacing per-department portals (IT, HR, finance, legal, procurement, workplace)
• Natural-language routing to the right AI Agent for the request
• Persistent conversation memory across sessions and devices
• Channel deployment to web, Slack, Teams, mobile app, and SMS
• Identity-scoped context — each employee sees their own records, approvals, history
• Integration with Workflow Data Fabric for cross-source answers ("show me my devices, my apps, my approvals")

Limitations: Replaces per-department portals only when the underlying departments are *also* on ServiceNow — partial deployments still leave employees dealing with a hodgepodge. Adoption is the harder problem than configuration: portals get muscle memory and EmployeeWorks needs visible behavior change to displace them. Conversational quality is dependent on AI Agents quality, which is dependent on Workflow Data Fabric grounding.

Best for: Understanding the entire architecture — Now Platform is the underlying app dev / workflow / data platform that *all* of ServiceNow's products run on. Every product (ITSM, HRSD, CSM, Field Service, Finance, Legal, Procurement, App Engine) is a Now Platform application. The April 2026 Foundation / Advanced / Prime retiering applies at the platform tier, not per-product.

• April 2026 tiers, marketed under the Autonomous Workforce umbrella brand: Foundation (AI assistance: summarization, recognition, categorization), Advanced (AI completes part of the work, agentic workflows alongside humans, Process Mining), Prime (fully autonomous workforce, agentic specialists, BYO-LLM)
• All tiers embed Now Assist, AI Control Tower, and Workflow Data Fabric — no separate purchase
• Token pools per tier with overage fees for AI consumption beyond pool
• App Engine for custom Now Platform applications; Build Agent for IDE-out workflow development
• Integration Hub for non-ServiceNow connectors; Now Platform Studio for low-code app creation
• ESM Foundation: midsize-company bundle bringing IT, HR, legal, finance, procurement, workplace onto one Now Platform tenant

Limitations: Pricing is the most opaque part of the stack — list prices for Foundation / Advanced / Prime are not publicly disclosed; ServiceNow says new tiers are 'set lower than what customers would have paid separately' but actual quotes still come through sales and are heavily NDA-bound. Token pool sizing matters — undersized pools turn the 'AI is included' promise into surprise overage bills. Lock-in risk is high: once your IT / HR / finance backbone is on Now Platform, switching cost is enormous.