Now Assist

What It Is

Now Assist represents ServiceNow's integration of generative artificial intelligence directly into the platform's core workflows, powered by the proprietary Now LLM (Large Language Model). Unlike standalone AI tools or external integrations, Now Assist operates as an embedded capability that leverages existing ServiceNow data, user context, and business logic to generate intelligent responses within familiar platform interfaces. The system doesn't replace existing ServiceNow functionality but augments it by providing AI-generated summaries of incidents, suggesting next steps for problem resolution, generating case descriptions, and offering contextual recommendations based on historical platform data. This differs from general-purpose AI tools like ChatGPT or Claude because Now Assist understands ServiceNow's data model, respects access controls, and generates responses specifically formatted for platform consumption.

Architecturally, Now Assist sits at the presentation and application logic layers of the ServiceNow stack, functioning as an intelligent interface enhancement rather than a fundamental data or integration component. The system processes requests through ServiceNow's existing API framework, ensuring that AI-generated content respects domain separation, access controls, and field-level security just like any other platform interaction. Now Assist capabilities are delivered through specific UI components, form sections, and workflow integrations that can be enabled or disabled at the application, table, or field level. This architectural approach means Now Assist inherits the platform's scalability, security model, and upgrade path rather than requiring separate infrastructure or introducing new technical debt. The AI processing occurs within ServiceNow's cloud infrastructure, maintaining data residency and compliance requirements that enterprises depend on.

From a business operations perspective, Now Assist addresses the chronic knowledge management and efficiency challenges that plague ITSM implementations. Service desk agents spend significant time reading through incident histories, searching knowledge bases, and determining appropriate next steps—work that becomes exponentially more difficult as ticket volumes increase and institutional knowledge walks out the door with departing employees. Now Assist generates contextual summaries of complex incident chains, suggests resolution steps based on similar historical cases, and helps agents craft clear communications without requiring deep technical writing skills. In ITOM contexts, Now Assist can analyze alert patterns and suggest correlations that human operators might miss during high-stress outage situations. For ITAM processes, the AI can generate asset descriptions, suggest categorizations, and identify potential compliance issues based on configuration patterns it has learned from the existing CMDB.

ServiceNow developed Now Assist as a response to customer demands for intelligent automation that doesn't require complex implementation projects or specialized AI expertise. Traditional AI implementations in enterprise software required separate platforms, data integration projects, model training, and ongoing maintenance that most ServiceNow customers couldn't realistically support. By embedding AI directly into the platform and training models on ServiceNow-specific use cases, ServiceNow eliminated the integration complexity while ensuring that AI suggestions align with ITIL processes and ServiceNow best practices. This approach contrasts with alternatives like integrating external AI services through REST APIs or building custom AI solutions, both of which introduce security, compliance, and maintenance overhead that enterprises typically struggle to manage effectively. The decision to use a proprietary LLM rather than integrating with public models like GPT reflects ServiceNow's focus on data privacy and compliance requirements that prevent many enterprises from sending internal data to external AI services.

End users interact with Now Assist primarily through enhanced form interfaces and suggested actions that appear contextually within their normal workflows—they might see AI-generated incident summaries, suggested knowledge articles, or draft responses that they can edit before sending. Platform administrators control Now Assist through system properties, application settings, and role-based permissions that determine which AI features are available to different user groups and in which contexts those features activate. Developers interact with Now Assist through APIs and configuration options that allow them to customize AI prompts, integrate AI-generated content into custom applications, and control how AI suggestions are presented within custom interfaces. Process owners typically engage with Now Assist through reporting and analytics that show how AI suggestions impact resolution times, user satisfaction, and process efficiency. Each group's interaction model reflects their different needs: end users want transparent assistance that makes their jobs easier, administrators need granular control over feature deployment, developers require integration flexibility, and process owners need measurable business impact.

Without Now Assist, ServiceNow instances would continue to function normally, but organizations would lose access to contextual intelligence that can significantly improve user productivity and service quality. Service desk teams would continue to manually review incident histories, search knowledge bases, and craft responses from scratch—processes that work but consume time that could be spent on higher-value activities. Problem management would lose AI-generated pattern recognition that can identify correlations between seemingly unrelated incidents. Change management would miss AI-suggested impact assessments based on historical change data. Most critically, organizations would lose the ability to scale their service operations efficiently as request volumes grow, because traditional manual processes don't scale linearly with staff additions. The absence of Now Assist doesn't break ServiceNow functionality, but it represents a missed opportunity to leverage institutional knowledge embedded in the platform's data to improve service delivery outcomes.

Where It Fits in the Platform

Now Assist integrates into ServiceNow's existing application framework as a cross-cutting capability that enhances multiple applications rather than functioning as a standalone module. The AI features appear within familiar ServiceNow interfaces—incident forms, knowledge article creation, case management, and problem investigation workflows—rather than requiring users to navigate to separate AI-specific applications. This integration approach means Now Assist inherits the platform's existing security model, data access patterns, and user experience paradigms while adding intelligent capabilities to established business processes. The system leverages ServiceNow's existing APIs, notification framework, and workflow engine to deliver AI-generated content through the same mechanisms that deliver other platform functionality.

The AI capabilities connect to ServiceNow's broader ecosystem through the same integration patterns that other platform features use—REST APIs for external system integration, event management for triggering AI analysis, and the notification system for delivering AI-generated insights to relevant stakeholders. Now Assist respects domain separation boundaries, role-based access controls, and field-level security just like native ServiceNow functionality. This architectural consistency means that Now Assist features can be included in update sets, controlled through system properties, and managed using the same governance processes that control other platform capabilities. The AI processing occurs within ServiceNow's infrastructure but operates on data that flows through existing ServiceNow data pipelines, ensuring that AI insights reflect the same data governance and quality controls that apply to other platform operations.

Key Relationships

• Knowledge Management: Now Assist generates suggested knowledge articles based on incident patterns and can automatically populate article content from successful resolution steps. The AI analyzes knowledge base usage patterns to suggest relevant articles during incident resolution.
• Incident Management: AI-generated incident summaries analyze work notes, related records, and resolution steps to provide contextual overviews that help agents understand complex cases quickly. Now Assist suggests next steps based on similar historical incidents and current incident state.
• Problem Management: The AI identifies patterns across multiple incidents to suggest potential problem records and analyzes root cause investigation data to generate insights about systemic issues. Now Assist can correlate seemingly unrelated incidents based on technical details and timing patterns.
• Workflow Engine: Now Assist integrates with ServiceNow workflows to trigger AI analysis at specific process steps and can generate workflow activities based on AI-recommended actions. The AI recommendations can automatically populate workflow variables and influence routing decisions.
• Role-based Access Controls: AI feature availability and the scope of data that Now Assist can analyze are controlled through standard ServiceNow roles and access controls. Users only see AI-generated insights for records and data they have permission to access through normal platform security.
• Configuration Management Database: Now Assist analyzes CMDB relationships and configuration data to suggest impact assessments for changes and identify potential service dependencies that might be affected by incidents. The AI can generate asset descriptions and suggest CI relationships based on discovered infrastructure patterns.

How You Encounter This in Practice

Service Desk Agent Reviewing Complex Incident

A Level 2 service desk agent receives an escalated incident that has been active for three days with fifteen work note entries, two assignment group transfers, and multiple related records including a change request and three child incidents. Instead of spending twenty minutes reading through the complete incident history to understand the current state, the agent sees a Now Assist-generated summary that highlights the key troubleshooting steps attempted, identifies which solutions were unsuccessful, and surfaces the most recent technical findings. The AI summary also suggests three potential next steps based on similar incidents that were successfully resolved in the past six months. Understanding Now Assist's capabilities allows the agent to quickly orient themselves to the problem context and focus their expertise on evaluation and execution rather than information gathering. Without this knowledge, the agent would manually review all work notes and related records, potentially missing important connections between related incidents or overlooking successful resolution patterns from similar historical cases.

Platform Administrator Planning Now Assist Deployment

A ServiceNow administrator receives executive pressure to "implement AI" across their ITSM processes but needs to balance innovation with risk management and user adoption. They discover that Now Assist features can be enabled selectively by application, user role, and even specific tables, allowing for controlled pilot deployments rather than enterprise-wide activation. The administrator realizes that Now Assist respects existing domain separation configurations, meaning they can test AI features in development domains without affecting production data or user experience. They also learn that Now Assist generates audit logs and usage analytics that demonstrate business value and identify potential issues before broader rollout. Understanding these deployment options allows the administrator to create a phased implementation plan that demonstrates AI value while maintaining platform stability and governance standards. Without this knowledge, the administrator might attempt an all-or-nothing deployment that overwhelms users, creates support issues, or fails to deliver measurable business outcomes that justify the investment.

Developer Integrating AI into Custom Application

A ServiceNow developer building a custom asset management application needs to incorporate intelligent asset categorization and risk assessment features without building machine learning capabilities from scratch. They discover that Now Assist APIs allow custom applications to request AI-generated content by providing context data and specific prompts, receiving formatted responses that can be integrated directly into custom forms and workflows. The developer learns that Now Assist maintains the same security context as the requesting user, ensuring that AI-generated insights only reflect data the user has permission to access. They also find that AI requests can be configured with custom prompts and response formats that align with their application's specific business logic and user interface requirements. Understanding these integration capabilities allows the developer to enhance their custom application with intelligent features that feel native to the ServiceNow experience while leveraging enterprise-grade AI infrastructure. Without this knowledge, the developer would either build complex custom AI integrations that require specialized expertise and ongoing maintenance, or deliver applications that lack the intelligent capabilities that users increasingly expect from modern enterprise software.

What People Get Wrong

Now Assist generates suggestions and insights based on patterns in historical ServiceNow data, but like all AI systems, it can produce plausible-sounding content that contains factual errors, outdated information, or inappropriate recommendations for specific contexts. The AI doesn't understand business nuance, regulatory requirements, or organizational policies that might make a technically correct suggestion operationally inappropriate. Users who treat AI-generated incident summaries, suggested resolution steps, or recommended knowledge articles as authoritative without verification can make decisions based on incomplete or incorrect information that leads to service outages, security vulnerabilities, or compliance violations. This misconception often develops because Now Assist content is presented within trusted ServiceNow interfaces using the same visual design patterns as verified platform data, creating an implicit assumption of accuracy that doesn't match the AI's actual capabilities.

The reality is that Now Assist functions as an intelligent research assistant that processes large amounts of ServiceNow data to surface relevant patterns and generate contextual suggestions, but human expertise remains essential for validating, customizing, and implementing AI recommendations. Effective Now Assist deployment requires establishing review processes where subject matter experts evaluate AI suggestions before implementation, particularly for high-impact activities like problem root cause analysis, change impact assessments, or security incident responses. Organizations that skip this validation step often experience incidents where AI suggestions led to incorrect troubleshooting approaches, inappropriate change approvals, or missed critical service dependencies. The business impact can be severe: a major retailer experienced a three-hour outage when service desk agents followed an AI-suggested resolution procedure without verifying that the suggested database restart would affect their e-commerce platform during peak shopping hours.

This misconception persists because AI-generated content often appears comprehensive and authoritative, especially when it incorporates specific technical details from incident histories or knowledge base articles that users recognize as accurate. The AI's ability to synthesize information from multiple sources creates an impression of thorough analysis that can override users' natural skepticism about automated recommendations. In production environments, this leads to over-reliance on AI suggestions without appropriate validation processes, inadequate documentation of AI-assisted decisions, and insufficient fallback procedures when AI recommendations prove incorrect. The solution requires treating Now Assist as a powerful analytical tool that enhances human decision-making rather than replacing human judgment, with clear governance processes that define when AI suggestions require additional verification and who bears responsibility for validating AI-generated content before implementation.

Now Assist uses ServiceNow's centrally-managed Now LLM rather than implementing organization-specific machine learning that adapts based on individual instance data or user feedback. The AI model was trained on ServiceNow-relevant data during development and doesn't update or improve its responses based on how users interact with AI suggestions in production environments. While Now Assist generates contextual responses by analyzing current data from your ServiceNow instance, it doesn't learn from whether users accept, reject, or modify AI suggestions, nor does it become more accurate by observing successful outcomes from AI-recommended actions. This architectural approach means that Now Assist capabilities remain consistent across ServiceNow instances and software updates, but it also means that the AI doesn't develop specialized knowledge about your organization's unique processes, terminology, or business context over time.

This misconception typically develops because other AI tools and platforms often emphasize machine learning capabilities that improve through usage, creating an expectation that enterprise AI systems automatically become more accurate and relevant through interaction. Users expect that repeatedly correcting AI suggestions or providing feedback on AI-generated content will result in better future recommendations, but Now Assist doesn't implement these learning mechanisms. Organizations that operate under this assumption often neglect to establish proper AI governance processes because they expect the system to self-correct problematic patterns, fail to maintain consistent AI prompt configurations because they assume the system will adapt to changing requirements, and don't invest in proper training because they believe users will naturally guide the AI toward better performance through daily usage.

The practical impact of this misunderstanding includes persistent AI accuracy issues that organizations expect to resolve automatically but require manual configuration changes or process adjustments to address effectively. Teams waste time providing detailed feedback on AI suggestions without realizing that this feedback doesn't influence future AI behavior, leading to frustration and reduced confidence in AI capabilities. More critically, organizations may deploy Now Assist features expecting that initial accuracy issues will resolve through usage, but without proper configuration and governance, AI suggestion quality can actually degrade as data quality issues accumulate or business processes evolve beyond the AI model's training scope. Success with Now Assist requires treating it as a sophisticated tool that requires ongoing configuration management, user training, and process optimization rather than an adaptive system that improves automatically through organizational usage patterns.

Admin vs Developer Perspective

For Admins

Admins control Now Assist through the sys_ai_config table and the AI Configuration module, where they enable or disable specific AI capabilities per application and user role. They manage data visibility boundaries by configuring which tables and fields Now Assist can access, making decisions about sensitive data exposure that directly impact compliance and security. The critical admin decision is balancing AI usefulness with data governance — allowing too much access creates audit risks, while restricting too much makes the AI suggestions useless. Admins also monitor AI usage through the sys_ai_audit table to track costs and identify users who might be over-relying on AI assistance.

For Developers

Developers integrate Now Assist into custom applications using the sn_ai scoped API, particularly the AIAgent class for generating text and the AISummarizer class for creating contextual summaries. They build custom AI prompts by creating records in the sys_ai_prompt_template table and can override default AI behaviors by implementing custom AI search sources or result filters. The key development pattern is wrapping AI calls in try-catch blocks since LLM responses can fail unpredictably, and always providing fallback logic when AI suggestions aren't available. Developers also need to understand token limits and implement chunking strategies for large content summarization.

How It Connects to Other Concepts

• Virtual Agent — Now Assist powers Virtual Agent's natural language understanding and response generation, replacing the previous intent-based conversation flow model. When users ask complex questions, Virtual Agent leverages Now Assist to search knowledge articles and generate contextually relevant answers rather than matching predefined patterns.
• Knowledge Management — Now Assist automatically generates article summaries and suggests related content by analyzing the kb_knowledge table content and user search patterns. It also helps authors improve articles by suggesting missing information based on common user queries that don't find satisfactory answers.
• Case Management — Now Assist analyzes case descriptions and work notes to suggest next steps, recommend similar resolved cases, and auto-generate status updates. It uses historical resolution patterns from the sn_customerservice_case table to predict likely solutions and estimate resolution times.
• Predictive Intelligence — Now Assist complements traditional ML models by providing natural language explanations for predictions and recommendations. While Predictive Intelligence identifies patterns in structured data, Now Assist interprets unstructured text content and generates human-readable insights about why certain predictions were made.
• Text Analytics — Now Assist builds on Text Analytics sentiment analysis and classification by adding generative capabilities that can create summaries and suggested responses based on identified text patterns. Where Text Analytics categorizes content, Now Assist generates new content informed by those categories.
• Performance Analytics — Now Assist generates natural language explanations of PA dashboard data and trends, making complex metrics accessible to non-technical users. It can analyze indicator breakdowns and suggest process improvements based on performance patterns, essentially providing AI-powered commentary on your PA data.

Junior vs Senior Knowledge Gap

Junior administrators typically treat Now Assist as a simple on/off switch, enabling it globally and assuming it will automatically improve user experience across all applications. They often miss the critical data governance implications, particularly around sensitive fields in HR, financial, or security-related tables that could leak through AI responses. The common mistake is enabling AI search across entire table schemas without considering that AI-generated summaries might expose information that users shouldn't see directly, even if they have read access to the underlying records. Juniors also tend to ignore the token costs and usage patterns, not realizing that poorly configured AI features can generate significant cloud computing expenses.

The mental shift happens when professionals realize that Now Assist isn't just a user interface enhancement — it's a data access and processing layer that needs the same security and governance controls as any other system integration. Experienced architects understand that AI suggestions are only as good as the underlying data quality and that garbage data creates confidently wrong AI responses, which are often more dangerous than obvious system errors. They know to implement AI features gradually, starting with low-risk use cases like knowledge article summaries before enabling AI-generated suggestions for critical business processes.

Senior professionals know that Now Assist performance depends heavily on prompt engineering and context management, not just the base LLM capabilities. They understand that the AI's effectiveness varies dramatically based on data volume — it works well with rich historical data but provides weak suggestions for new processes or edge cases. They've learned to build manual override capabilities into any AI-driven workflow because LLMs can fail in ways that traditional code doesn't, producing responses that seem reasonable but are factually incorrect or contextually inappropriate. The experienced approach involves extensive testing with real data, not just demo scenarios.

The questions that separate senior architects include asking about AI model versioning and backward compatibility — what happens when ServiceNow updates the underlying LLM and changes response patterns that users have grown to expect? They inquire about data residency and processing location for AI requests, particularly important for organizations with strict compliance requirements. They also ask about AI decision auditability and how to trace AI-generated content back to its source data for regulatory purposes. Most importantly, they design AI implementations with clear boundaries and fallback mechanisms, treating generative AI as an enhancement to human decision-making rather than a replacement for business logic.

Quick Reference

• Now Assist has a 32,000 token limit per request, roughly equivalent to 24,000 words of context — summarization requests that exceed this fail silently and return generic responses
• The sys_ai_audit table tracks every AI interaction but doesn't store the actual generated content for privacy reasons, only metadata like token count and response time
• AI search respects ACLs at query time but may include information from restricted fields in generated summaries if the user has any read access to the record
• Now Assist requires the sn_ai_core plugin and consumes monthly AI credits that reset on your instance's contract anniversary date, not calendar months
• Custom prompt templates in sys_ai_prompt_template can reference GlideRecord variables and system properties but cannot execute server-side scripts or access external APIs directly
• AI-generated case summaries automatically exclude work notes marked as work_notes_list.type = 'internal' but may include internal comments from other activity fields
• Virtual Agent conversations using Now Assist maintain context for 30 minutes of inactivity, after which the AI forgets previous conversation history
• The AIAgent.generate() method has a 5-second timeout in synchronous mode but can run up to 30 seconds when called asynchronously through scheduled jobs
• Now Assist knowledge search indexes only published articles with workflow_state = 'published' and ignores draft or retired content, even if users have direct access to those records
• AI-generated text in form fields doesn't trigger standard business rules or data policies until the user manually saves the record, preventing accidental workflow automation from draft AI content