How should ServiceNow rethink its workflow thesis for AI buyers?

ServiceNow has to evolve from "workflow orchestration platform" to "enterprise AI agent control tower" — and the window is roughly 18 months before the narrative ossifies. The four mental shifts: (1) workflows running *on* the platform → agents running workflows *as* the platform, (2) ticket-to-resolution → agent-to-resolution-without-ticket, (3) Pro Plus as an AI uplift SKU → Pro Plus as the agent platform itself, (4) "single platform of record" → "agent platform of record *plus* the system of record underneath." The one narrative bet that risks the moat: going too aggressively agent-first can orphan the ITIL/ITSM IT-ops buyer who built the install base — Bill McDermott has to thread *agent-first language for the AI buyer* without telling the legacy CIO their Now Platform purchase was a stepping stone.

The bet is real because Salesforce (Agentforce), Microsoft (Copilot Studio), and pure-play agent infra (Decagon, Sierra, Lindy, Cresta) are all racing for the same control-tower position, and ServiceNow's Workflow Data Fabric / RaptorDB is the strongest defensible substrate nobody is talking about loudly enough.

The Old Workflow Thesis (what got ServiceNow to $11B ARR)

• Single platform of record — one CMDB, one workflow engine, one data model across ITSM, HRSD, CSM, FSM
• ITIL-grounded — ticket → assignment group → resolution → CMDB update is the atomic unit; the entire org chart maps to this
• IT-buyer trust — the CIO is the economic buyer, procurement loves the consolidation story, and the ROI math is "retire 12 point tools"
• Pro / Pro Plus seat-based pricing — predictable per-user uplift, easy to forecast, sales comp aligned
• Named legacy references — Deloitte, NVIDIA-as-customer, Visa, AstraZeneca, US DoD — all bought "workflow consolidation," not "agent platform"

What AI Buyers Actually Want In 2026

• Agent-first, not ticket-first — the work unit is an autonomous agent loop, not a Jira/ServiceNow ticket that a human closes
• Model-flexible, not vendor-locked — buyers want to swap Claude / GPT / Gemini / open-weights without re-architecting; "bring your own model" is table stakes
• Data-portable, not platform-trapped — Iceberg / Delta / open table formats, not proprietary Now-only data graves
• Outcome-priced, not seat-priced — pay per resolved case, per booked meeting, per closed incident — not per logged-in human
• Audit-ready agent governance — every agent action logged, evals on every release, EU AI Act + SOC2 alignment, human-in-the-loop policy controls
• Composable, not monolithic — the AI buyer wants to assemble agents from MCP servers and best-of-breed components, not buy a 7-year platform

The 4 Mental Shifts ServiceNow Needs

• Shift 1 — From "workflow runs on platform" → "agents run workflows on platform." The atomic unit changes. A workflow used to be a deterministic state machine triggered by a ticket. Now it's an agent reasoning loop that *invokes* deterministic workflows as tools. ServiceNow has to reframe Now Assist + AI Agent Studio as the substrate where agents *live*, not as a feature bolted onto Flow Designer.
• Shift 2 — From "ticket → resolution" → "agent → resolution-without-ticket." The aspirational metric stops being MTTR on a ticket and becomes "% of would-be tickets that never opened because an agent resolved upstream." Deflection rate at the edge — not after the fact.
• Shift 3 — From "Pro Plus = AI uplift SKU" → "Pro Plus = agent platform." Today Pro Plus is sold as a 30-60% price uplift for AI features. The new pitch: Pro Plus *is* the agent platform; classic Pro is the legacy compatibility tier. Flip the hierarchy.
• Shift 4 — From "platform of record" → "agent platform of record + record platform." Two-layer pitch: ServiceNow is the *agent* control tower (where agents are governed, observed, and orchestrated) AND the underlying record platform (CMDB, HR record, customer record). The agent layer is the new top — the record layer is the moat underneath, not the marketing lead.

The 1 Narrative Bet That Risks The Moat

• The 18-month repositioning cliff — go too hard on agent-first and the ITIL/ITSM CIO who renews 40% of ServiceNow's ARR feels orphaned; go too soft and Salesforce Agentforce + Microsoft Copilot Studio + Sierra/Decagon eat the agent narrative
• Named risk #1: ITSM stalwarts — Deloitte, Accenture, big-bank ITOps teams bought "single platform of record"; if the new pitch is "agent control tower," those CIOs need a migration story, not a re-pitch
• Named risk #2: Pro Plus pricing inversion — if Pro Plus becomes "the agent platform," classic Pro looks like a legacy SKU and renewals get harder; sales comp has to be re-architected before the narrative ships
• Named risk #3: AI Agent Studio cannibalizing Flow Designer — the agent platform partially obsoletes the no-code workflow tool that Now built its developer ecosystem on; the partner channel revolts if not handled
• The hedge — McDermott's framing has to be additive ("the agent layer sits on top of the platform of record you already own") not replacive ("forget tickets, agents now")

What McDermott's Pitch Should Sound Like

• Investor-day language — "ServiceNow is the control tower for enterprise AI agents — the only platform where agents are governed, observed, and connected to the system of record." Drop "workflow OS" by Q3 FY26; lead with "agent OS, workflow native."
• Sales-pitch language — "You already own the system of record. We turn it into the system of *action* for every AI agent in your enterprise — yours, ours, and your vendors'."
• Product-page hierarchy — Now Assist + AI Agent Studio above ITSM/HRSD/CSM in the nav, not below. Workflow Data Fabric promoted from "data layer" to "AI context substrate."
• Named-customer reference patterns — stop leading with "Deloitte consolidated 12 ITSM tools"; start leading with "NVIDIA runs 200 production agents on the ServiceNow agent platform with full audit trails and model-swap flexibility." Same logos, new headline metric.

What Has To Change Operationally

• Now Assist becomes flagship product, not a feature — its own P&L, its own GM, its own launch cadence; out of the Flow Designer org
• AI Agent Studio becomes the primary GTM motion — sales is comped on agent-platform attach rate, not Pro Plus seat uplift
• Workflow Data Fabric / RaptorDB repositioned as AI context substrate — Iceberg compatibility, vector-native, MCP-server native; the "why ServiceNow under your agents" moat
• Named acquisition target — Decagon (CX agent infra) or Lindy (long-tail agent builder) to bolt agent-first DNA onto the org and signal commitment; alternatively, Cresta for the contact-center agent angle
• Open up the model layer — public BYO-model support for Claude / GPT / Gemini / Llama with first-party benchmarks; kill the implicit "Now LLM is best on Now" narrative
• Agent governance becomes a paid module — Agent Observability + Eval + Policy as a standalone Pro Plus capability; this is the audit-ready governance layer the EU AI Act mandates

Old Thesis × New Thesis × Mental Shift × Cost × Risk × Timeline

Old Thesis → Pivot → New Thesis

FAQ

How should ServiceNow reposition its core narrative for AI buyers? ServiceNow has to evolve from "workflow orchestration platform" to "enterprise AI agent control tower," with a roughly 18-month window before the narrative ossifies. The strongest defensible substrate is Workflow Data Fabric / RaptorDB, which the article argues nobody is talking about loudly enough.

The hedge is keeping the record platform (CMDB, HR record, customer record) as the moat underneath the new agent layer.

What are the four mental shifts ServiceNow needs? The four shifts are: workflows running on the platform → agents running workflows as the platform; ticket-to-resolution → agent-to-resolution-without-ticket; Pro Plus as an AI uplift SKU → Pro Plus as the agent platform itself; and "single platform of record" → "agent platform of record plus the system of record underneath." Shift 2 changes the aspirational metric from MTTR to the percentage of would-be tickets that never opened.

What do AI buyers actually want in 2026? AI buyers want agent-first rather than ticket-first work units, model-flexible "bring your own model" support to swap Claude, GPT, Gemini, or open weights, and data portability via Iceberg/Delta/open table formats rather than Now-only data graves.

They also want outcome pricing (per resolved case, per booked meeting) over seat pricing, audit-ready agent governance aligned to the EU AI Act and SOC2, and composable agents assembled from MCP servers.

What is the one narrative bet that risks the moat? Going too aggressively agent-first can orphan the ITIL/ITSM IT-ops buyer who built the install base and renews ~40% of ServiceNow's ARR. Named risks include ITSM stalwarts like Deloitte and Accenture needing a migration story, Pro Plus pricing inversion making classic Pro look legacy, and AI Agent Studio cannibalizing Flow Designer and angering the partner channel.

The hedge is McDermott framing the agent layer as additive ("sits on top of the platform you already own"), not replacive.

What should McDermott's investor-day pitch sound like? The recommended language is "ServiceNow is the control tower for enterprise AI agents — the only platform where agents are governed, observed, and connected to the system of record." The article says to drop "workflow OS" by Q3 FY26 and lead with "agent OS, workflow native." The product-page hierarchy should put Now Assist and AI Agent Studio above ITSM/HRSD/CSM, promoting Workflow Data Fabric from "data layer" to "AI context substrate."

Bottom Line

ServiceNow's workflow thesis got it to $11B ARR; it will not get it to $20B. The pivot is real, the substrate (Workflow Data Fabric + RaptorDB + the system-of-record gravity) is genuinely defensible, and McDermott has the narrative chops to land it — but the 18-month window is unforgiving and the Pro Plus pricing inversion has to be sequenced before the agent-first marketing flips.

Bet: ServiceNow makes the pivot but only after a named acquisition (Decagon / Lindy / Cresta tier) forces the org to internalize agent-first DNA. *(see also: q1613, q1614, q1620)*