While the industry was abuzz in 2024 with the term 'Generative AI', 2025 will be the year for AI Agents. A recent LangChain research survey revealed over half of companies already use AI agents, with over 75% planning to do so in 2024. Are you considering yourselves part of this demographic? If not, I highly recommend reading and watching the content in this blog.
AI Agents represent a transformative leap beyond traditional ML, Virtual Agents & GenAI.
Here's why this is such a important step forward:
1.) The Ability to Reason + Act Autonomously
Agentic AI, or AI Agents, refers to artificial intelligence systems designed to act autonomously and proactively in a goal-directed manner. AI Agents are capable of making decisions, taking actions, and interacting with their environment without continuous human intervention, powered by LLMs, access to tools, and a specific algorithm around observing, planning, reasoning, and acting. This makes them far more scalable and adaptable to complex, undefined workflows, which are the best types of problems for the technology. AI Agents can operate autonomously, working together through a virtual orchestrator to complete tasks without human supervision. This capability includes the ability to interface with multiple systems/tools to gather data and take action. While there's still a gap between this AI reasoning and human reasoning, and while there is debate on the nature of the reasoning vs reciting, it has proven to improve performance and achieve autonomous goal achievement powered by LLMs.
2. Reduced Implementation Complexity for Unknowns
One of the historical challenges with Virtual Agents and static workflows has been the costly and time-consuming implementation process to define decision trees, intents, entities for all permutations of a users inputs. While there are definitely run-time cost (compute) and latency benefits to using decision trees or static defined workflows, AI Agents eliminate the need for these rigid decision trees, significantly reducing deployment barriers and improving time-to-value in solving open ended problems. As a LLM based solution, it also means the system can do tasks like classification without requiring a large golden data set for training (e.g. determine category of an issue based on available categories). Given a goal, it will dynamically create a plan and consume available tools, utilizing natural language as an input from users to direct a goal and clarify any followup action.
The Shift Toward AI Agents
The fundamental role of SaaS has been to provide UIs & databases to simplify workflows and tasks for knowledge workers to enhance employee productivity/ customer revenue. As AI evolves, up to 15-20% of future knowledge worker roles tasks are becoming handled by AI Agents & GenAI, allowing for additional capacity to be freed for greater productivity. In many cases, AI Agents are being viewed as the next engagement & logic layer of the enterprise, accessing the needed data and taking the necessary (CRUD) actions on behalf of the users, and aligned to organizational policy and access controls. They benefit from their natural language style input to synthesize and present information. With this initial context on AI Agents set, I'd like to share a brief informal overview of our AI Agent plans for ITSM:
SaaS providers like ServiceNow grew in popularity because it offered operational cost savings & value differentiation over on-prem solutions or tool sprawled alternatives. The next frontier of cost efficiency lies in autonomous AI agents augmenting your workforce, built a...
Why to Consider ServiceNow ITSM AI Agents
Our ownership of workflows, system of record, and domain context positions us uniquely to build domain-specific agents out of the box. Unlike competitors, we can deliver integrated, purpose-built AI Agents optimized for ITSM. These agents can access data directly from the instance, also invoking integrations, flows, scripts, and tools. In addition, we can define the access scope of the user we want the agent to run as- meaning you can maintain your existing security and data definitions established on the platform. In addition, ServiceNow houses many of your organizations business logic, such as approvals, categorization, routing, prioritization, escalation, etc. The AI Agents can take advantage of this nuanced business logic, data, and full workflow orchestration for end to end transformation to improve agent productivity and end user employee experience through faster MTTR via process automation and automated issue resolution.
In terms of the out of the box agents, ServiceNow is targeting a variety of use cases from generic issue resolution to process and data quality optimization. While some more static use cases like specific IT issues (password reset) may be better suited by LLM workflows that predefine tools and logic, since the workflow is very immutable and known, the AI agents excel at the generic issue handling through unknowns. These AI Agents can be triggered manually via the now assist panel, or set to run autonomously when a record is created/updated. You can also define what actions require a supervised human in the loop to confirm.
You can watch a demo of the technology in action below, as well as watch a full ITSM Agentic AI presentation from our latest Knowledge 25 conference.
Our AI agents take advantage of industry standard agent architecture infused with our data, business logic & tools. Specifically, we use a multi-agent architecture that is differentiated from a single agent approach in its ability to improve quality and accuracy, at the cost of latency. These individual agent specialists get invoked and managed by a orchestrator that ensures achievement of plan and defined goals. As a platform, you may see future capabilities for monitoring your AI Agents performance, impact, and any issues with their setup (AI Agent Observability), as well as tools to automate the testing and validation of your new AI Agent workforce.
Individual AI Agents are a combination of perception, memory, reasoning, planning and action. They do so using an AI agent architecture that involves an agent framework, a memory (which tracks past conversation), access to tools, and a planning system. Memory itself is a incredible benefit because it allows AI Agents to gather context from a users feedback in a unstructured manner, that can then impact future replies. Traditionally, data needed to be structured and placed in a table. Now, unstructured feedback can be remembered to change replies and behavior.
And since they are built on the AI Platform of ServiceNow, that means you still get all the experience layer engagement points for employees and agents, alongside other AI capabilities like AI Search. You can see a demo of how AI Search & AI Agents can augment requesters and IT agents below.
So What's Next? The Strategic Future for AI In ITSM
ITSM’s Objectives
The core objective of ITSM can be viewed as two outcomes for the organization:
Improve organizational productivity, which can be measured and supported by:
- Employee CSAT & self service experience
- Reduced MTTR and count of incidents
- Empower digital transformation across org
Optimized business operations, through cost saving and revenue generation:
- Run IT efficiently at minimum operating expenses (opex/capex)
- Support business objectives, especially revenue generating services
- Improving service quality and availability
Traditionally, CIOs and ITSM leaders in organizations scaled to improve these outcome via investments in:
- Specialized IT Knowledge Workers
- IT Software to structure workflows/data required by the knowledge workers
- Physical hardware required to run the software/services
- Vendors that provided above services
AI is revolutionizing IT Service Management, dramatically enhancing organizational productivity, streamlining business operations, and significantly reducing operational costs. ITSM continues to move towards a 0 incident, fully automated service desk. Users issues are resolved in near real time, self service engagement is improving, and services have 100% uptime due to proactive analysis and prevention.
AI’s Role in ITSM
With the continual improvements in AI in terms of model accuracy, capability, and cost, we are seeing a shift in how ITSM leaders can achieve their intended outcomes of productivity and efficiency, without having to rely on the traditional investment economics.
In the past, AI was used to solve a specific step of a workflow – routing a user’s question (NLP), categorizing a field (supervised ML), recommending a knowledge article (Unsupervised/Similarity/AI Search), or generating resolution notes (GenAI). This saved time and improved data quality within an existing human-in-the-loop process, and AI was merely a tool for augmenting worker productivity in specific steps of the process.
The major shift we have encountered in recent months is the ability of AI, through LLMs’ AI Agent reasoning, to string together the steps and take action for complete task management, from inception to resolution. This means a traditional incident can be triaged, routed, diagnosed, resolved, and wrapped up with knowledge written entirely autonomously by an AI agent with access to the same tools and information that a human agent would use. A change can be planned and analyzed based on past data and industry practices. A problem can be identified from a series of incidents, the root cause can be examined, and a solution can be proposed and even implemented. Every core ITSM workflow across incident, problem, change, and request will be augmented by AI. This augmentation comes in two forms: insight and action.
AI for Insights
Identifying trends/patterns in data like repeated incidents
Recommending risk of a change
Generating resolution notes or major incident status updates
AI for Action
Proactive changes, tasks, and alerts
AI Teammates – owning an end-to-end workflow from record creation to closure
Digital twins – an AI agent with all the context of your files, conversations, and preferences
These two forms of AI apply to the two forms of data and interactions generated in ITSM – structured and unstructured. Structured data represents the records in the system like incidents with specific fields like CI’s and Categories, whereas unstructured data represents emails, chat messages, and freeform text. AI can now produce and synthesize both types of ITSM data.
Core Changes in AI This Year
One of the bigger shifts this year is the move to supporting agentic AI in the form of ‘digital employees’ or AI teammates. Traditional knowledge work includes specialized domain experts manipulating information and data, making decisions and reasoning on context to take actions. Traditionally, AI only influenced the available information, or took actions based on predetermined decision criteria. With todays Agentic AI, we now see LLMs handling the reasoning and decision making step as well. This means the entirety of a knowledge work process, especially where actions are all software based, can be automated.
Core ITSM roles will have digital AI Agents with the same role description and instructions to follow the same processes as the human employees have today - AI Service Desk Agent, AI Network Engineer, AI Major Incident Manager. These AI teammates can be assigned automatically or invoked manually by todays agents to support them in their ITSM workflows. A part of this shift will be the creation of the level 0 agent – a new AI agent level that will handle initial triage, intake, and attempted resolution or recommendation for teams. As a result, the work existing L1 agents will handle will naturally increase in complexity, where additional ai recommendations can help them upskill without experience. AI is no longer solving a step in a workflow for the human, AI is owning the workflow calling the human to solve steps as needed.
Another notable change is the shift in engagement layer, driven by the improvements in LLM understanding and generation of text and speech, linked to the capability of AI Agents to act and reason. Specifically, we are seeing requestor engagement layers continue to move to chat & voice, with users specifying their needs in natural language as opposed to learning to navigate a UI. Requestors will also see more proactive engagement on their devices from device level AI agents that can guide them through what they are seeing in real time and User interfaces in general will simplify as AI can be more dynamic and personalized in content displayed. Depending on a incidents state (e.g. diagnosis, resolution, or post incident review), the content on the screen will display key insights relevant to that phase. Fulfillers will be able to interact conversationally with a UI to gain insights on data, and
Future trends
As ITSM leaders continue to use AI to improve the efficiency of core ITSM workflows, we will see the next evolution being a move to solve productivity and business objectives directly. There will be a move from solving employee productivity reactively (e.g. ensuring employee remains productive through available services, hardware, etc.), to solving productivity directly – offering GenAI tools to help knowledge workers generate reports, insights, etc. As AI autonomy improves and AI gets closer to general artificial intelligence, the need to manage and track digital AI workers in organizations will grow. IT agents will need UI to work alongside their AI teammates, seeing work that needs their input and their performance. This will be a new concept of an AI hub or AI cockpit, where human fulfillers can see the work of their ai agents, any escalations where they need help, and their current work and performance. In this regard, every IT fulfiller will virtually manage their own peer of AI agents, stepping in where needed or providing guidance/feedback for resolution.
With the rapid rise in adoption of GenAI & AI Agents, IT will grow an increasing role in the management of these AI technologies and deployments. AI adoption is the next phase of digital transformation for many line of businesses, supported by IT. This includes the need to manage all AI investment (models, applications) across vendors in a AI Control Tower, to ensure risk and compliance and reduce unneeded technology spend. At the same time, IT will begin monitoring the performance of models and AI teammates to ensure accuracy and satisfaction through the same reporting and monitoring used in observability systems and regular worker management dashboards. There will be an introduction of AI Agent Service Levels, with triggers defined to escalate AI Agent to a human based on time, number of actions/LLM calls, or user sentiment.
The control tower will be more important when we start to see the maturation of the AI software factory, where new AI deployment opportunities are identified by AI, a solution generated, deployed to production, then monitored and altered based on performance KPIs and user feedback. IT teams will also maintain more agent to agent integrations over standard REST APIs to exchange data and take actions across systems.
As AI continues to automate routine process and work, IT can shift investment to more strategic improvement initiatives and creative innovation. IT will move from reactive to proactive and have a larger percent time spent on upgrading infrastructure, code, processes, as opposed to responding to incidents. We will also see the gradual shift from ITSM resources towards ITOM, as more incidents will be machine generated and less human generated. This may evolve into a generalist IT staff as most low priority and routine incidents on both sides become automated.
Conclusion
CIO ITSM organizations will see augmentation in core ITSM workflows from incident to change via insights and actions from AI. Organizations will be able to improve core employee productivity and business efficiency metrics without significant investment due to AI advancements in AI’s ability to increasingly augment and take on ‘digital labor’. Employees will improve their self serve experiences via natural language engagement channels, while fulfillers will have access to AI colleagues to handle tasks autonomously. The productivity improvements for the fulfillers and agents will carry down to requestors as general productivity assistants. Improved self service, continuous monitoring and always on AI Agents will slash MTTR and improve service availability. The rise of these AI deployments will lead to the need for clear monitoring and governance from IT. The ITSM of the future will rely heavily on the management and improvement of these AI colleagues. The productivity benefit of these autonomous AIs will also allow IT to become more strategic with more time spent on improvement initiatives like upgrading software and infrastructure, as well as being more proactive in delivering services from a ITOM lense.
All the best to 2025,
Darius
ITSM Product Management
Safe harbor statement: The article above is intended to outline our general product direction. It is intended for information purposes only, and may not be incorporated into any contract. It is not a commitment to deliver any material, code, or functionality, and should not be relied upon in making purchasing decisions. The development, release, and timing of any features or functionality described for ServiceNow’s products remains at the sole discretion of ServiceNow.
