Starting with the Brazil release, NLU topic discovery in ServiceNow Virtual Agent is being prepared for future deprecation. Virtual Agent itself stays, and LLM topics become the remaining topic discovery option. As of 22 October 2026, ServiceNow has published no removal date. This guide covers what the migration tool converts, what it leaves for you, how to retest discovery, and what each topic costs in Assists. It also maps Now Assist in Virtual Agent to the Otto brand that replaced Now Assist.
- No end date: NLU stays supported and will be hidden on new instances, per ServiceNow's Brazil NLU documentation.
- Copy, then fix: the Migrate Topics to LLM workflow copies NLU topics, and dynamic topic blocks migrate only when you select them by hand.
- Cost: ServiceNow's Now Assist Overview prices one end-to-end Virtual Agent topic execution at 10 Assists.
Is ServiceNow deprecating Virtual Agent or only NLU topic discovery?
ServiceNow is preparing NLU topic discovery for deprecation and keeping Virtual Agent. The Brazil Virtual Agent release notes say Virtual Agent for NLU and Virtual Agent Lite "are being prepared for future deprecation." They "will eventually be hidden and no longer activated on new instances but will continue to be supported." No removal date is published.
ServiceNow NLU deprecation reaches past Virtual Agent. The same Brazil release notes mark the Virtual Agent NLU functionality in com.glide.cs.chatbot for deprecation with the instruction "Migrate NLU topics to LLM." The same notes list NLU Workbench-Advanced Features, Intent Discovery, and Topic Recommendations as planned for deprecation with no replacement. They name Requestor Agents - Foundation and EmployeeWorks - User as replacements for Virtual Agent Lite. The Brazil deprecation summary adds ITSM Virtual Agent pre-built topics to the list. Echelon's post on upgrade surprises that catch teams off guard includes reviewing deprecation notices as upgrade prep.
The change starts with the Brazil family. The Zurich deprecation summary and its Australia counterpart list no NLU items. Zurich Patch 12 already shows the direction inside the instance: Conversational Studio 11.0.4 displays a banner in the Assistant Designer Asset library when NLU assets are present, per the August 2026 Now Assist suite release notes.
Brazil is in early availability. A ServiceNow staff post gave the Brazil early access date as a tentative 10 September. A later staff post said Brazil early availability moved to 24 September. ServiceNow's docs give no exact GA date. A ServiceNow staff member put the target at "GA early November" in a June 2026 Community reply. Plan the migration before your Brazil upgrade lands. If the same upgrade includes a workspace move, see moving from Agent Workspace to Service Operations Workspace.
Is ServiceNow Otto for Virtual Agent the same as Now Assist in Virtual Agent?
ServiceNow Otto for Virtual Agent is the current name for Now Assist in Virtual Agent, the LLM layer that runs on the Virtual Agent plugin. ServiceNow's Zurich Patch 12 release notes call Otto "the new AI experience brand" and state "Your product entitlements remain unchanged." The plugin ID still carries the old name.
ServiceNow announced Otto at Knowledge 2026 on 5 May 2026 as a new AI experience combining Now Assist, Moveworks, and AI Experience. The press release names EmployeeWorks and AI Control Tower as the first surfaces and does not mention Virtual Agent. In the instance, Virtual Agent is the Glide Virtual Agent plugin (com.glide.cs.chatbot), and ServiceNow Otto for Virtual Agent is the sn_nowassist_va plugin, per ServiceNow's Exploring Virtual Agent page.
The rename is uneven. The October 2026 suite release notes still list the Store app as "Now Assist in Virtual Agent," version 20.0.8 for Patch 11. A ServiceNow staff Otto and Now Assist quick start guide says the switch-over "will be uneven." Older runbooks, Community threads, and KB articles use both names for the same product.
The prerequisite wording changed between releases. The Zurich Otto for Virtual Agent notes say it "requires a license for Virtual Agent and at least one Now Assist product." The Brazil release notes say it "requires at least one ServiceNow Otto product or the Requestor Agents - Foundation plugin." Confirm which wording applies to your release with your account team. Renewal questions about your Assist pool are covered in how Assist pools work under the 2026 tiers.
ServiceNow NLU vs LLM: how does LLM topic discovery pick a topic?
LLM topic discovery in ServiceNow matches what a user types against each topic's plain-language description. NLU discovery matches against trained intents and utterances. ServiceNow's LLM topic discovery documentation says LLM topics need no models, intents, or keywords. When several topics match, the user picks one from a list.
ServiceNow's discovery documentation limits topic switching: it works during a query and stops inside a catalog ordering flow, per the Zurich LLM topic discovery page. Each LLM topic must also be associated with an assistant to be discoverable, according to ServiceNow's Virtual Agent readiness guide.
NLU and LLM topics are separate estates. A ServiceNow staff article states that NLU topics "can't run in the Now Assist chat client, and LLM topics can't run in the NLU-based Virtual Agent," in Building Virtual Agent topics that answer user questions. ServiceNow's KB2388191 on Otto for Virtual Agent says NLU topics "often require months of development" and that existing NLU topics can keep running while you migrate them.
ServiceNow also claims a quality edge for LLM topics. Its Xanadu LLM topic discovery page says conversations that use LLMs "can outperform" NLU conversations. ServiceNow publishes no benchmark behind that claim, so measure discovery accuracy on your own utterances before cutover.
How should you write LLM topic descriptions?
Write descriptions as instructions to the LLM. ServiceNow's LLM description and instruction guidelines say "Keywords aren't necessarily helpful, but logic is." The Zurich migration documentation warns that "NLU utterances themselves are not effective if you use them as topic descriptions for LLM topics." Keep each topic to five user input controls or fewer, per ServiceNow's input control guidance.
What does the Migrate Topics to LLM tool convert?
Migrate Topics to LLM is the ServiceNow workflow that copies NLU and keyword topics into new LLM topics and leaves the originals unchanged. It writes descriptions with generative AI, maps entities and vocabulary sources, and records what it could not convert in a migration log. Dynamic topic blocks migrate only when selected by hand.
For a Virtual Agent NLU to LLM move from Zurich onward, you migrate NLU topics to LLM from All > Conversational Interfaces > Assistant Designer, on the Asset library tab, per the Zurich procedure to migrate NLU topics. Users with virtual_agent_admin or sn_vad_genai.topic_migration_admin can run it. The tool first shipped "As of Washington Patch 3," according to a 2024 ServiceNow staff article on migrating NLU topics. That article's opening sentence names keyword-based topics, while its note says "Keyword topics are not supported for migration." The Xanadu documentation on topic migration later says all topic types migrate.
The workflow runs in six steps: select topics, review connections, choose settings, migrate, review descriptions, and publish, per the Yokohama migration workflow steps. The connections step covers topic blocks, notifications, and Proactive Triggers. The settings step includes a "Keep current topic descriptions?" option that is off by default, so generative AI writes new descriptions unless you change it. The review step can test each new description against the original NLU topic's utterances.
| Item | What the tool does | What you do | ServiceNow source |
|---|---|---|---|
| Topic copy | Creates new LLM topics. The original NLU and keyword topics stay unchanged. | Decide when to deactivate the NLU originals after cutover. | Yokohama topic migration concepts |
| Descriptions | Generates descriptions from up to 10 utterances, set by sn_vad_genai.utterances.count.for.topic.description. | Rewrite each description as logic and instructions. | Australia topic migration concepts |
| Entities and vocabulary | Adds entity details to each node's detail description, with content that differs slightly by entity type. List vocabulary sources become static choice nodes; table sources become dynamic choice nodes. | Check extraction on every input node. | Xanadu topic migration concepts |
| Switch-topic scripts | Logs a switch topic script issue when a switch topic function was not migrated. | Select the referenced topics on the Select topics step, then retest the script. | NLU to LLM migration log |
| Dynamic topic blocks | Migrates them only when you select them by hand. | Select each block on the Topic blocks tab of the Review connections step. | Yokohama migration procedure |
| Script and data-pill prompts | Replaces them with a "Collect + Node name" placeholder. | Write a detail description for each node. | Yokohama node description rules |
| Empty descriptions | Migrates the topic with an empty description, which must be filled in before publishing. | Add a description before publishing. | Zurich empty description rule |
| Notifications and Proactive Triggers | Connects them only to topics published through the workflow. The Proactive Triggers tab needs sn_pt 3.0.1 or later. | Publish through the workflow's publish step. | Zurich publish step requirements |
| Resume | Loses your place if you leave the workflow. | Finish each batch in one session. | Australia migration procedure |
| Batch size | ServiceNow advises "fewer than 50 topics at one time." | Size batches to what your team can review. | Zurich migration procedure (linked above) |
How do you read the NLU to LLM migration log?
The NLU to LLM migration log is the table topic_migration_execution_item, also downloadable as a CSV from the workflow. ServiceNow's Xanadu migration log reference sorts issues into node-level, script-level, topic-level, LLM-generated description, NLU intent, dynamic topic block, switch topic script, and process failure categories. Work the switch topic script and dynamic topic block categories first: both involve topics the workflow did not migrate. The 2024 staff article adds one irreversible case: "If you migrate a read-only topic, you can't delete it!"
Should you convert a topic or rebuild it?
ServiceNow publishes no criteria for choosing between converting a topic and rebuilding it. Its ServiceNow Otto for Virtual Agent readiness guide tells admins to use the topic migration tool to convert NLU topics to LLM. It offers ServiceNow's pre-built ITSM LLM topics "as a starting point for Virtual Agent topic creation." The signals below are judgment built from documented limits.
Signals we look for, each tied to a limit ServiceNow documents:
| Signal | Documented limit | ServiceNow source |
|---|---|---|
| More than five inputs in one topic | ServiceNow advises against placing more than five user input controls in a single LLM topic. | LLM input control limits |
| Catalog ordering that relies on topic switching | Topic switching stops inside a catalog ordering flow. | Australia LLM topic discovery page |
| Input nodes built on NLU entity properties | The NLU entity property is available only for NLU topic discovery. | Input control guidance (linked above) |
| Heavy switch-topic scripting | The migration log flags switch topic functions that were not migrated. | Migration log issue types |
| Dynamic topic blocks | Dynamic topic blocks must be selected by hand. | Manual dynamic block selection |
A topic that trips two or more signals is a candidate for a rebuild from its business intent. Zurich Patch 12 also added "Bulk migration of Topics to AI Agents" in Conversational Studio 11.0.4, per the same Zurich Patch 12 Store release notes, which adds AI agents as a migration target. Scope any rebuild as a pilot first, using running an AI POC on ServiceNow.
What should you retest after migrating ServiceNow Virtual Agent LLM topics?
Retest every migrated topic in the live client your users touch. Practitioners have reported a topic that works in the Designer test console and fails in the live Now Assist client, as in a March 2026 Community report on migrated topics. ServiceNow documents interactive testing in Assistant Designer.
- Test live-client discovery: run utterances in Now Assist on Service Portal as well as the console, as ServiceNow staff advise for discovery testing.
- Check description overlap: ServiceNow staff say "Overlapping descriptions are what produce the ambiguous list in the first place," in a staff troubleshooting article.
- Confirm fallback and AI Search: the AI Search Fallback setup topic fires when discovery finds no match; without AI Search, staff say fallback answers do not generate.
- Cover every channel: ServiceNow's Otto for Virtual Agent KB lists the chat widget, the mobile app, and Microsoft Teams.
- Test each language: Dynamic Translation is supported for non-streaming conversations, per the Zurich to Brazil Otto for Virtual Agent notes.
- Trace switch-topic calls: filter the migration log for switch topic script issues and confirm each script switches to an LLM topic.
- Watch Assist usage in sub-production: Now Assist in Virtual Agent cannot run on a personal developer instance, and ServiceNow's Now Assist Overview counts sub-production use against your Assists.
Practitioners report other LLM topic failures in ServiceNow Community threads. One topic was discoverable as admin but not when impersonating a user. The stock My Approvals topic worked in the standard chatbot and failed with an access error as an LLM topic. A Zurich assistant built from scratch dropped straight to live-agent routing. None of the three threads has an accepted solution.
ServiceNow documents ATF for Virtual Agent (com.glide.cs.atf) on a page titled Automated testing for Virtual Agent topics that use NLU topic discovery. ServiceNow does not document ATF for LLM topic discovery. The Create test case button works only on published topics that are not read-only, per ServiceNow's ATF test creation page.
Interactive testing fills the gap. The topic test window on the designer canvas has an Analysis tab with topic discovery results, documented on the test details reference. ServiceNow's testing LLM topics page says to raise low scores by updating the topic description or the instructions in LLM user input nodes. Keep the rest of the regression pack current with ATF best practices for regression coverage, and see how AI cuts ATF test build time.
How many Assists do ServiceNow Virtual Agent LLM topics use?
ServiceNow prices one end-to-end Virtual Agent topic execution at 10 Assists in its legal Now Assist Overview, effective 24 August 2026. A Now Assist panel conversation costs 5. Search costs 0, or 5 per result when Quick GPT or World Knowledge results are included. AI Search answer cards cost 0. ServiceNow publishes no rate for LLM topic discovery alone.
| Skill in the Now Assist Overview | Counted action | Assists consumed |
|---|---|---|
| Now Assist Virtual Agent topics | One end-to-end Virtual Agent topic execution | 10 |
| Now Assist panel conversation | One conversation, ending after 1 hour of inactivity or a change in intent | 5 |
| Search (knowledge, forms, password reset portal deflection) | One plugin execution | 0, or 5 per result with Quick GPT or World Knowledge |
| Now Assist Q&A genius results | One answer card in AI Search results | 0 |
| Multi-context synthesized answer generation | One answer generated | 0 |
| Subflows and actions | One end-to-end conversational flow execution | 10 |
| Agentic workflow (small, medium, large) | Less than 4, 5-8, or 9-20 actions | 25, 50, or 150 |
The ServiceNow Virtual Agent cost model counts executions. The Overview has no per-conversation rate for Virtual Agent and no row for a discovery turn that ends without a topic. A ServiceNow staff article says Virtual Agent topics running as tools inside an agentic workflow are billed as part of the workflow run. The Overview also states that Now Assist skills consume Assists in production and sub-production instances, so migration testing draws on the same pool. Echelon's breakdown of how Now Assist assists are metered walks through sample rate-card rows.
Track consumption at Admin > Subscription Management > Subscriptions > Now Assist usage, with the usage_admin, sn_sub_man.admin, or admin role, per ServiceNow's Now Assist usage monitoring page. The per-skill table on that page updates nightly. Subscription Management v6.0.2 moved consumption to a 365-day burn-down model in January 2026. Set threshold rules in AI Admin Hub usage alerts, and read totals per assistant on the Assistant analytics Assists dashboard.
Where do Employee Slate and EmployeeWorks fit with Virtual Agent?
Employee Slate is an AI-first employee experience bundled inside ServiceNow EmployeeWorks and powered by Moveworks or Now Assist. Its Now Assist mode does not support Virtual Agent NLU topics, and the Employee Slate product FAQs recommend LLM topics. ServiceNow has published nothing on Virtual Agent topics under Employee Slate powered by Moveworks.
ServiceNow renamed the Moveworks version to EmployeeWorks Web App in September 2026, and the Now Assist version keeps the Employee Slate name. The Moveworks timeline sets the context. ServiceNow announced the Moveworks acquisition in March 2025 for $2.85 billion and completed it on 15 December 2025. EmployeeWorks became generally available on 26 February 2026, and the same release said Moveworks continues to be offered as a standalone product. Employee Slate followed on 5 May 2026. Its first release required Zurich Patch 9, per a ServiceNow staff launch post. The October 2026 release requires Zurich Patch 13, Australia Patch 7, or Brazil Patch 1, per the EmployeeWorks Web App implementation guide.
Virtual Agent appears inside Employee Slate as an assistant. With the com.snc.ex_ai_portal app installed, Assistant Designer shows a default Employee Slate assistant that comes with premium chat, per the Zurich to Brazil release notes for Otto. Employee Center Pro stays: the Employee Slate FAQ says "there are no plans to deprecate Employee Center Pro." The Employee Slate FAQ on Assists also says Employee Slate consumes Assists, because it is predominantly a conversation-based experience. A separate plugin, Moveworks for Employee Center, replaces the standard Virtual Agent with Moveworks conversational AI in Employee Center. To plan a move from Employee Center to Employee Slate, see planning an Employee Slate migration. For Arc, ServiceNow's personal AI agent that is now part of EmployeeWorks, see what ServiceNow Arc is.
Frequently asked questions.
ServiceNow Otto vs Now Assist: what changed?
ServiceNow Otto is the brand that combines Now Assist, Moveworks, and AI Experience, announced on 5 May 2026. ServiceNow's Otto product page says Otto replaces Now Assist and Moveworks across all ServiceNow products, channels, and surfaces. Zurich Patch 12 release notes state that product entitlements remain unchanged. ServiceNow said in February 2026 that Moveworks continues to be offered as a standalone product.
ServiceNow Virtual Agent vs Now Assist: are they separate products?
Yes. Virtual Agent is the base conversational product, and Now Assist in Virtual Agent, now ServiceNow Otto for Virtual Agent, is a separate Store app that adds LLM topics. In the instance they are separate plugins: com.glide.cs.chatbot and sn_nowassist_va. The Brazil activation requirements say it needs at least one ServiceNow Otto product or the Requestor Agents - Foundation plugin.
Can NLU and LLM topics run in the same assistant?
ServiceNow documents them as separate. A ServiceNow staff article says NLU topics cannot run in the Now Assist chat client and LLM topics cannot run in the NLU-based Virtual Agent. KB2388191 says existing NLU topics can keep running while you migrate them. One practitioner in a January 2026 Community thread wrote that, in their understanding, the LLM assistant has no NLU fallback.
Has ServiceNow published a date when NLU stops working?
No. The Brazil release notes say NLU in Virtual Agent is being prepared for future deprecation and will continue to be supported, with no removal date. ServiceNow points readers to its Deprecation Process article (KB0867184) for how the phases work. The Zurich and Australia deprecation summaries contain no NLU items.
Should a new Virtual Agent build use NLU or LLM topics?
New builds on Brazil should use LLM topics. ServiceNow says NLU in Virtual Agent will be hidden and no longer activated on new instances. ServiceNow's Brazil deprecation summary lists Virtual Agent for NLU and Virtual Agent Lite as being prepared for future deprecation. Brazil's Exploring Virtual Agent page names LLM discovery as the remaining topic discovery option.
How do you run AI Search from an LLM topic?
ServiceNow has not documented an AI Search block for LLM topics. Practitioners asked about it in a February 2026 Community thread that remained unresolved in September 2026. The documented path is the AI Search Fallback setup topic, which fires when topic discovery finds no matching topic. Staff note that fallback answers need AI Search on the instance.
Can you test LLM topics on a personal developer instance?
No. ServiceNow staff confirmed in February 2025 that Now Assist is not available on PDIs, so LLM topic testing needs a sub-production instance with Otto entitlements. ServiceNow's Now Assist Overview counts sub-production use against your Assists. Plan the test cycle with the complete guide to ServiceNow ATF testing for the NLU-era tests you keep.
How does Echelon help with a Virtual Agent migration?
Echelon is the AI workforce for ServiceNow delivery. It handles requirements gathering, architecture, coding, testing, documentation, and deployment, per Echelon's AI ServiceNow developer launch. Build output lands as an update set in a sub-production instance for your developer to review, and Echelon generates ATF coverage for the change, as described in AI tools for ServiceNow delivery compared.
To scope a Virtual Agent topic estate before your Brazil upgrade, book a scoping call.
