AI Agents in IT: From Hype To Hands-on Impact

Organizations are scrambling to implement AI in the enterprise to boost productivity, increase efficiency and gain competitive advantage. In 2024, corporations invested $252.3 billion in AI, yet the impact is mixed: While most organizations see positive financial impacts, most reported cost savings of less than 10% and revenue increases below 5%.*

Enterprises must be able to transform AI projects from abstract promises into tangible gains that give business users what they need when they need it. Part of that means building applications that are more resilient and robust, with better throughput, error reduction and automated remediation, because they are AI-enabled. AI agents also need to be integrated throughout the software development life cycle (SDLC), including in CI/CD pipelines, observability platforms and incident response tools.

There are many areas where organizations are integrating AI into IT operations, including incident remediation, predictive monitoring and test generation, supporting improved mean time to detect (MTTD) and mean time to resolution (MTTR) and helping ensure application uptime and performance. A recent report found that by initially focusing AI tools on the applications used in IT operations, organizations can demonstrate value, obtain a swift return on investment (ROI) and gain confidence in how to direct their future AI efforts more broadly across the enterprise.

How Agentic AI Can Reshape IT Operations Workflows

IT workflows and use cases are probably the ripest for agentic AI. These tools are designed to help the business stay online and fulfill its objectives by correlating massive amounts of information, signals and events. The goal is to detect issues and address them before they impact performance and the business.

Further, those working in IT operations are accustomed to automation. Unlike many business users who may be reticent to turn functions over to agents, IT staffers are more likely to rapidly gain confidence in what agents are doing. Hence, AI has a smoother path to adoption in IT operations than in the enterprise as a whole.

By starting small within IT to prove agents out and scaling them across IT operations, the value of AI agents can quickly become apparent. Built upon that strong foundation, agents can be applied to business-focused tasks with a greater level of expertise.

After all, IT personnel are the ones developing and implementing AI. Getting them on board by demonstrating tangible benefits in the everyday tasks they perform is a vital buy-in step. If IT professionals are enthusiastic about rolling out agentic AI into the enterprise, and they are fully supported in their efforts by the C-suite, it becomes much easier to obtain buy-in among business units.

Key Elements of an Agentic AI Framework

When planning agentic AI initiatives, it’s wise to establish a framework to help assess where AI agents are most likely to bring improvements, how to evaluate AI readiness in different parts of the organization, how to manage and orchestrate the actions of AI agents, and how to implement governance for responsible oversight.

Key elements of such a framework could include:

IT ops and cybersecurity

• Automating best practices in IT operations and cybersecurity.
• Automatically remediating issues and enabling further automation with automated workflows.
• Automating compliance checks with compliance-as-code recommendations, reducing audit risk and regulatory exposure, and creating audit-ready artifacts.
• Generating zero-configuration dashboards, alerting, troubleshooting and remediation.
• Giving applications the precise resources they need when they need them, including optimizing GPU workloads without sacrificing performance.
• Identifying resource congestion and cost overruns across hybrid cloud environments.

Data-based recommendations

• Scoring and correlating data to identify issues and implement actionable recommendations.
• Consolidating scattered data so it can provide insights and generate actions to prevent problems, not just provide alerts.
• Providing actionable recommendations, thresholds and remediation plans using agentic AI.
• Displaying evidence in visualizations and natural language interactions so that users can validate findings, explore context and move to remediation.

Software development

• Scanning for and detecting problematic code that can result in security or resilience issues.
• Automating full-stack application visibility across the entire monitoring life cycle, including real-time change detection, mapping, tracing and profiling.
• Enabling automatic and continuous discovery, deployment, configuration and dependency mapping.

Traditionally, these tasks would require hopping from screen to screen to review CPU, memory and disk metrics, checking log files to gather more data, and drilling down in the search for the root cause. Instead, agents can automate all of this and correlate actions between different agents to recommend actions, or if they are fully trusted, take specific actions based on preset policy.

A user can begin by reviewing an evidentiary trail to see why agentic AI recommended a specific action. A human can then approve the action or adjust it as appropriate. As a greater percentage of the IT operations workload becomes automated, the user can review why an AI agent did what it did. In this way, the system can be continually streamlined and improved. Those skeptical of agentic AI can gradually gain confidence in its findings and decrease the degree of supervision. This frees up IT to engage in more strategic actions.

AI agents can also be molded to fit the needs, priorities and preferences of the enterprise or specific applications. In some cases, performance will be king, while in others, cost savings will dominate. The agents can be adjusted to enterprise requirements, create policies about specific tasks and take prescribed actions in response to exact situations. The next time the system encounters a known scenario, it can automatically perform its duties. Depending on the application and the business case, a human can be completely in the loop, completely out of the loop, or can step in and out based on their level of comfort and trust.

In the end, it is all about providing the right type and amount of resources, in the right place, at the right time. Underprovisioning resources slows teams down, whereas overprovisioning, especially expensive GPU compute resources, is a serious waste of money.

AI Agents in the Real World: Success Stories

There are many examples of AI agents being used to automate tasks in IT and business operations and provide tangible value. Here are some examples.

• One site reliability engineering (SRE) team used AI agents to unlock a standardized and repeatable approach to application resilience. Results: Reduced person-days in an IBM enterprise-wide resilience posture evaluation per application by 62%, compared to manual evaluation; reduced the IBM SRE team’s person-hours spent in monthly operations review by 72% compared to manually producing the report; and by breaking down silos, IT became more productive. Additionally, the enterprise put a scalable framework in place to measure, improve and sustain application resilience across the organization.
• In a benchmark challenge, AI agents helped accelerate common vulnerabilities and exposures (CVE) management at scale. Result: 90% faster CVE mitigation and 98% faster certificate inventory management.
• In a global financial services organization pilot program, AI agents accelerated patch management from 90 minutes to a maximum of 20 minutes per instance and reduced the “median time to patch” for critical vulnerabilities from 80 hours to 8 hours.

How To Start Small, Scale Responsibly

There is an ongoing race to use AI to achieve enterprise glory. Enterprises rushing headlong into AI may end up making mistakes that reduce confidence in agentic AI. The smart approach is to start small, preferably in IT operations use cases, and scale responsibly within IT, before unleashing automated agents across the business.

IBM Concert, Instana and Turbonomic are designed to provide the framework required for rapid ROI with agentic AI. They provide integrated application observability and risk management that can help unlock the potential of your applications with AI-driven automation backed by a comprehensive governance framework.

Learn more about how IBM Observability can simplify complexity and scale resilience.

* Nestor Maslej, Loredana Fattorini, Raymond Perrault, Yolanda Gil, Vanessa Parli, Njenga Kariuki, Emily Capstick, Anka Reuel, Erik Brynjolfsson, John Etchemendy, Katrina Ligett, Terah Lyons, James Manyika, Juan Carlos Niebles, Yoav Shoham, Russell Wald, Tobi Walsh, Armin Hamrah, Lapo Santarlasci, Julia Betts Lotufo, Alexandra Rome, Andrew Shi, Sukrut Oak. “The AI Index 2025 Annual Report,” AI Index Steering Committee, Institute for Human-Centered AI, Stanford University, Stanford, CA, April 2025. (CC BY-ND 4.0)