Five years of selling AI as a premium add-on, and most enterprise customers are still waiting for the return they were promised. ServiceNow president Amit Zavery put it plainly in this week's analyst briefing: the failure is architectural, not technological. Disconnected data, siloed workflows, and no consistent governance layer mean agents cannot act correctly even when the underlying model is capable. Today's announcements are the company's answer to that structural problem.

Zavery has met more than a thousand customers and partners in five quarters at the company. He said they raise the same three complaints: AI is not connected to the workflows that run the business, data is fragmented across systems that do not talk to each other, and nobody can show the board a return. ServiceNow's position is that fixing those complaints requires owning the layer where workflow execution, data connectivity, and governance intersect. That is what today's three announcements are designed to do.

Context Engine Is Where the Decade of Workflow Data Finally Pays Off

Agents fail at enterprise scale because they lack institutional memory. They know a policy exists. They cannot determine whether it applies to this employee, in this role, for this request, right now. Context Engine is ServiceNow's answer. It pulls together identity data from Veza, asset data from Armis, generative business intelligence from Pyramid Analytics, and data catalog capabilities from Data.world, then layers in decision tracing from the Traceloop acquisition. Every approval, resolution, and escalation enriches the graph. The system learns how this specific business actually makes decisions, and that accumulated understanding is what gets applied at runtime.

ServiceNow SVP John Aisien described it as capturing the why behind decisions, not just the what. The demo example was a business-class travel request. Normally that requires a policy lookup, a manager approval, and a precedent check, each a manual handoff. With Context Engine, the agent knows the employee's role, the applicable travel policy, and how comparable requests were handled before, then approves or routes in seconds without human intervention. The distinction Zavery emphasized: this is not an AI making a guess. It is an AI applying judgment grounded in live enterprise context.

The foundation is not being built from scratch. ServiceNow has been developing Service Graph for IT service management and the Knowledge Graph for over a decade. Context Engine extends that operational history into identity, assets, and AI decision tracing. At Knowledge 26 in May, ServiceNow says Context Engine will shift from informing decisions to learning from them continuously. Context Engine is currently in preview with select customers; broader availability timing was not disclosed.

Build Agent Skills Opens the Platform to Developers Who Never Used ServiceNow

Jithin Bhaskar, general manager for Creator Workflows and App Engine, ran the Build Agent Skills demo and made a point worth holding onto: vibe coding is an ignition switch, not a production solution. The Build Agent Skills software development kit encodes the ServiceNow platform into whatever agentic development environment a developer already uses: Claude Code, Cursor, OpenAI Codex, or any other integrated development environment. The demo walked through generating a travel request application in VS Code using Claude Code, then deploying it directly to ServiceNow Studio. Governance, role-based access controls, and the audit trail were generated automatically on deployment, not added afterward. Every agent built through the SDK gets registered in AI Control Tower as an asset, with version history and a kill switch.

ServiceNow is not charging developers for the build tooling itself. Enterprise customers receive 100 free Build Agent calls to start; personal instance developers get 25. Beyond that on-ramp, pricing was not disclosed. That matters because a Build Agent call is not defined publicly. An organization scaling prompt-to-deployment across multiple development teams has no way to model costs before committing to usage, which makes finance conversations difficult. ServiceNow told briefing participants that subscription manager will gain what-if scenario analysis for usage projection, but that capability is in progress, not yet shipped.

Code generated through the SDK outputs in Fluent, ServiceNow's new native agentic language, rather than generic JavaScript. Fluent replaces the XML-based file structures that defined the older platform and is designed specifically for agent interactions. Developers working outside ServiceNow's environment still produce output that only runs correctly inside it. Any organization evaluating Build Agent Skills should factor that dependency into the platform commitment decision before scaling adoption.

Bundling AI Into Every SKU Eliminates the Add-On Excuse

Standard, Pro, and Enterprise are gone. Foundation, Advanced, and Prime replace them, with AI included across every tier at no additional licensing step. Moveworks capabilities, Workflow Data Fabric access, AI Control Tower basics, and process mining are bundled throughout the full price list. The previous model required customers to buy Plus add-on SKUs to access AI features inside products they already licensed. Removing that friction matters for adoption curves, because the single most reliable predictor of AI shelfware is a purchasing step that separates the capability from the workflow.

Each tier maps to a level of autonomous execution. Foundation covers AI-assisted work where humans remain in the loop: summarization, categorization, routing. Advanced automates complete process flows end-to-end. Prime enables the AI Specialist construct, which ServiceNow describes as autonomous role replacement for level-one functions including IT service desk, HR service delivery, and customer service. Role replacement is reserved for Prime; the lower tiers do not include it.

The consumption model uses a synthetic unit called an Assist, introduced two years ago with Now Assist. Assists are fungible across workloads and scale with action complexity: small deterministic actions consume fewer, large autonomous actions consume more. Data fabric usage carries its own token meter. ServiceNow expects customers to use subscription commitments for predictability and Assists for scaling AI adoption incrementally.

Alongside the core tier restructuring, the Enterprise Service Management Foundation offering packages the platform for midsize organizations with a claimed 30-day deployment target, using implementation agents and AI-guided setup. ServiceNow positions this as competitive pressure on vendors like Atlassian that have historically owned the midmarket IT service management segment.

Anthropic as Default Engine, Not Just a Partnership Announcement

Claude is the default model powering Build Agent. ServiceNow has also deployed Anthropic Claude and Claude Code to its own workforce of more than 29,000 employees. The sales preparation use case cut seller prep time by 95 percent in testing, according to ServiceNow. That figure is vendor-reported and unaudited. The engineering productivity numbers using Claude Code were not quantified specifically.

ServiceNow maintains a multi-model strategy and also has a partnership with OpenAI for agentic automation and speech capabilities. The Anthropic relationship is positioned as best tool for the task, specifically where complex code generation and agentic workflow reasoning are the requirement.

What the Architecture Claim Requires CIOs to Decide

ServiceNow is asking enterprises to designate it the control plane for enterprise AI: the single layer where agents are orchestrated, governed, audited, and observed. For organizations where ServiceNow already runs the highest-volume operational workflows, that is a natural consolidation. For organizations where Salesforce, Microsoft 365, or SAP represent the operational center of gravity, Context Engine compounds on data that largely lives elsewhere, and its value proposition shrinks accordingly.

Context Engine builds intelligence from every decision processed on the platform. Run most of your consequential decisions through ServiceNow and the graph gets smarter with every action. Run them across five platforms and Context Engine becomes one more disconnected intelligence layer, adding governance overhead rather than reducing it.

Budget modeling is the near-term risk that deserves more attention than the briefing addressed. Organizations scaling AI adoption across IT service management, HR Service Delivery, and customer service simultaneously will draw on Assist tokens, Data Fabric tokens, and Build Agent calls concurrently. Three consumption meters running in parallel against a subscription baseline creates real cost projection difficulty, particularly while the what-if scenario analysis tooling in subscription manager remains unshipped.