Knowledge 2026 was the year ServiceNow stopped talking about AI pilots and started talking about running the enterprise on agents. Our team was there, and we have spent the months since putting the announcements to work on customer platforms. This is our sorting of them, by what you should do about each.
Do this quarter: switch on AI Control Tower
The AI Control Tower, introduced a year earlier as a visibility product, became a full governance framework with five parts: discover, govern, secure, observe and measure. Its scope now reaches beyond ServiceNow’s own agents to models, agents and workflows running on AWS, Azure, Google Cloud, SAP, Oracle and Workday, through around thirty new integrations.
The detail that matters for your budget: Control Tower is included in every tier of the new commercial model, Foundation, Advanced and Prime. Customers on older contracts should check their entitlement at renewal rather than assume it. The expanded version went into the Innovation Lab in May with general availability in August, so by the time you read this you can switch it on.
Do it before your next agent goes live, not after. Our agentic AI article explains why the order matters.
Learn the new name: Now Assist is ServiceNow Otto
The other announcement that touches every customer is a rename. Now Assist, the Moveworks products ServiceNow bought in December 2025 and the AI Experience layer are being combined into one assistant called ServiceNow Otto. It went live first in EmployeeWorks and AI Control Tower, the Now Assist name is being replaced in the product from the August 2026 AI release onwards, and the rename completes with the Brazil release. Licensing and entitlements do not change.
For most teams this is documentation and training work rather than a project, but do it, because “Now Assist” will keep appearing in old runbooks and plugin names for a while. We cover what changes in our Otto article.
Plan for it: Action Fabric and the MCP server
Action Fabric exposes ServiceNow’s workflow engine, across IT, HR, customer service, security, risk and app development, to external AI agents through a generally available Model Context Protocol server. The pitch is straightforward: whatever agent your people use, Claude, Copilot, a home-grown bot or ServiceNow’s own, the work it does should run through the platform, with its security model and audit trail.
We have been running MCP servers against ServiceNow for a while, for customers and for our own consultants, so we can say from experience what it changes and what it does not. It changes how fast you can give an agent safe access to ServiceNow. It does not change the need for clean data, scoped roles and a clear owner for every action the agent can take. Plan a short piece of work to define the roles and the allowed actions before you open the door.
Evaluate: the Autonomous Workforce
ServiceNow expanded its catalogue of pre-built “AI specialists” for IT, customer relationship management, employee services and, new this year, security and risk. These are agents with a defined job, shipped with the product, that you configure rather than build.
Our advice is to evaluate them use case by use case against the agent you would otherwise build. For common jobs such as incident triage or knowledge drafting, a shipped specialist on good data will beat a custom agent on poor data every time. For anything specific to your organisation, you will still build. Either way, the data assessment comes first.
Watch: Australia and Brazil
Australia, the first release under the new country naming, reached general availability in May 2026, and Brazil entered release preview in September. Two releases a year is now the steady rhythm, and the AI capabilities move fastest between them. If your upgrade process still takes a quarter, that is the thing to fix before you worry about which release you are on. We wrote up how we run two upgrades a year.
What to ignore for now
Anything that depends on data you do not have. Most of the demos at Knowledge ran on a CMDB that was complete, a knowledge base that was current and a service map that was accurate. If yours are not, the right first project is not an agent. It is a two-week assessment of the data the agent will read, and then the agent.
If you want to work through the announcements against your own roadmap, that is a conversation we have most weeks. Talk to us.
