ServiceNow Otto: One Permission-Aware AI Layer

TL;DR: The ServiceNow Otto unified AI experience control tower is a single conversational layer that spans the whole enterprise instead of living inside one app. ServiceNow built it to fix the “completion problem” of enterprise AI: assistants that answer questions but cannot finish work because they lack the permissions, approval chains, and cross-system reach real tasks need. The signal for everyone else is plain. The demand is for one permission-aware layer across all tools, not a separate bot per application.

Most enterprise AI today can tell you the answer but cannot do the thing. You ask a tool a question, it responds well, and then you still open three other apps, chase an approval, and route your own request. ServiceNow gave that gap a name at its Knowledge 2026 conference and shipped a product against it.

On May 5, 2026, ServiceNow introduced ServiceNow Otto, a unified AI experience that combines Now Assist, Moveworks, and its AI Experience layer into one front door for employees, customers, and partners (ServiceNow Newsroom, 2026). The interesting part is not the brand name. It is what a company this size is betting on: that the next layer of enterprise AI is horizontal and permission-aware, not bolted into each separate product.

What is the enterprise AI completion problem?

The completion problem is the gap between answering and finishing. An assistant inside a single app can explain a policy or draft a reply, but it cannot push the work across the systems and approvals that actually close the loop.

ServiceNow describes the cause plainly. Other major software providers ship AI inside their own applications, “working in compartmentalized isolation, unable to complete work across departments or systems,” and large language models add intelligence but “don’t connect to a governed platform with the approval chains, permissions, audit trails, and cross-system workflows that enterprise work requires” (ServiceNow Newsroom, 2026). The result is AI that answers questions but cannot finish the job, while employees still toggle between applications and AI costs keep climbing.

That framing matters beyond ServiceNow. It names a problem most large organizations feel but rarely state: intelligence is now cheap, and connection is the bottleneck.

Why a single layer instead of AI in every app?

ServiceNow’s answer is to put the AI above the apps. “Rather than living inside a single application, ServiceNow Otto sits across the entire enterprise, understanding intent, routing work to the right agent, and executing it to completion” (ServiceNow Newsroom, 2026). The promise is that an employee no longer needs to know which system owns their request. They ask once.

Industry coverage read the move the same way. ServiceNow made a pitch that “whatever AI agent your employees end up using — Claude, Microsoft Copilot, a homegrown bot or one of ServiceNow’s own — the work those agents do should run through ServiceNow’s platform,” giving employees “a single front door” (Reworked, 2026). CIO reported the company’s own framing that “the era of sidecar AI is over” (CIO, 2026).

Strip away the branding and the architecture is the point. A per-app assistant can only see its own data and act within its own walls. A layer that spans tools can answer a question that touches HR, IT, and finance in one pass, because it can reach across all three.

The acquisition tells you how serious the bet is

This is not a weekend feature. Otto is built largely on ServiceNow’s purchase of Moveworks, a conversational AI and enterprise search company it bought for $2.85 billion in a deal that closed in December 2025 (Reworked, 2026). Moveworks supplied the front-end assistant and search; ServiceNow supplied the workflow backbone. Companies do not spend that to ship a chatbot. They spend it to own the layer employees talk to.

How does permission-aware governance fit in?

A conversational layer that spans every system is only safe if it respects who can see and do what. That is the role of AI Control Tower, the governance layer underneath Otto.

Actions taken through Otto are “governed by AI Control Tower, which can log each AI interaction, enforce enterprise policies, and provide explainability for every decision” (ServiceNow Newsroom, 2026). ServiceNow says Otto’s actions are “grounded in a customer’s data, policies, approval chains, and organizational structure” (ServiceNow Newsroom, 2026).

Grounded and permission-aware are the operative words. A horizontal assistant that ignored access rules would be a liability: it would either leak restricted data or stall the moment it had to act. The same logic runs through the broader Knowledge 2026 announcements, where the expanded Control Tower was positioned as the oversight layer for “every AI system running in an enterprise, not just the ones built on ServiceNow,” with the ability to flag and stop an agent that tries to act outside its permissions (Reworked, 2026).

Does grounded, cross-system AI actually get adopted?

Early numbers suggest people use AI when it finishes work rather than just talks about it. ServiceNow’s first surface for Otto, EmployeeWorks, “generated six deals exceeding $1 million each in net new annual contract value (NNACV)” just one month after launch, which the company offers as evidence “that when AI is grounded in enterprise context and completes real work, people adopt it” (ServiceNow Newsroom, 2026).

There is a structural reason this lands. ServiceNow reports more than 100 billion workflows running on its platform each year (ServiceNow Newsroom, 2026). An assistant sitting on top of that volume already has somewhere to send the work. Adoption follows completion, and completion follows connection. Analysts add the obvious caveat: a layer like this is only as good as the data under it. As one Forrester analyst put it, “agents reasoning at machine speed over a stale graph are going to produce wrong outputs” (CIO, 2026).

A concrete example: one question, three systems

Picture a regional hospital network, Vantage Health. A nurse manager, Priya, needs to onboard a traveling clinician starting Monday: provision a laptop, grant access to the scheduling system, and confirm the credentialing paperwork cleared compliance.

Today that is three tools and two emails. IT owns the laptop, the scheduling app owns access, and compliance owns the credential check. Priya has to know which system does what, file a request in each, and chase the approvals herself. The classic completion problem: every tool can answer “what’s the status,” none can finish the task.

A single permission-aware layer changes the shape of that work. Priya asks one question in plain language. The layer reads her intent, confirms she manages the unit and is allowed to initiate onboarding, and routes the laptop request to IT, the access grant to scheduling, and the credential check to compliance, each step logged and inside the existing approval chains. Priya does not learn three systems. She asks once and gets a result she can trust.

This is the same architecture SemanticOS is built around: a knowledge-graph and AI-search layer that connects fragmented enterprise tools so people, and AI agents, can find and reason over institutional knowledge across systems instead of inside one app. ServiceNow validating the horizontal, permission-aware model at this scale is a useful signal that the connective layer, not another point tool, is where enterprise AI is heading.

Key takeaways

• ServiceNow Otto is a unified AI experience that spans the whole enterprise, combining Now Assist, Moveworks, and AI Experience into one conversational front door (ServiceNow Newsroom, 2026).
• The target is the “completion problem”: AI that answers but cannot finish work because it lives inside a single app without cross-system permissions and approvals.
• A horizontal layer only works if it is permission-aware; AI Control Tower supplies the logging, policy enforcement, and explainability that keep actions governed.
• A $2.85 billion Moveworks acquisition and 100 billion annual workflows show the scale of the bet on a single layer over per-app assistants (Reworked, 2026).
• The wider lesson: enterprises want one connective, permission-aware layer across all tools, the model SemanticOS is built on, not isolated AI in every product.