The successes and challenges of AI agents

AI has changed a lot in just two years. In 2023, most companies were experimenting with large language models. These tools helped with writing, research, and support tasks. They were smart, but they waited for instructions and could not take action on their own.

In 2025, we are seeing something more powerful: AI agents. They are not just chat tools anymore. They can remember, plan, use tools, and act on their own. AI agents can take a broad goal, figure out the steps, and carry it out without needing help at every stage. Some can even fix problems along the way.

Early wins

These agents have moved beyond research and begun working inside real businesses. For example, ServiceNow uses AI agents to manage IT requests. If someone needs software installed or a license updated, the agent takes care of it from start to finish. There are no tickets to raise and no waiting time.

GitHub Copilot is another example. It now has a mode where the agent understands what the developer is trying to do, choosing tools, making decisions, and completing small coding tasks on its own. For developers, this saves time and removes repetitive work.

A final example is Cisco, which is using AI agents inside Webex to improve customer service. One agent speaks directly to customers, another supports human agents during live calls, and a third listens and creates a summary of the conversation with tone and sentiment analysis. These layers work together and make customer support faster and more accurate.

These applications of AI agents work well because the tasks are clear and follow a standard process. But agents are now being trained to handle more complex problems too.

Take this use case: A business analyst is trying to answer why sales dropped for a product last quarter. In the past, a human would explore the data, come up with possible reasons, test them, and suggest a plan. Now, an AI co-pilot is being trained to do most of that work. It pulls structured data, breaks it into groups, tests different ideas, and surfaces the insights. This kind of system is still in testing but shows what agents might be able to do soon.

A better approach

Even with these early wins, most companies are still trying to add agents to old workflows, which limits their impact. To really get the benefits, businesses will need to redesign the way work is done. The agent should be placed at the center of the task, with people stepping in only when human judgment is required.

There is also the issue of trust. If the agent is only giving suggestions, a person can check the results. But when the agent acts directly, the risks are higher. This is where safety rules, testing systems, and clear records become important. Right now, these systems are still being built.

One unexpected problem is that agents often think they are done when they are not. Humans know when a task is finished. Agents sometimes miss that. In some tests, over 30% of multi-agent failures were caused because one agent thought the task was completed too early.

To build agents, developers are using tools like LangChain and CrewAI to help create logic and structure. But when it comes to deploying and running these agents, companies rely on cloud platforms. In the future, platforms like AWS and Google Cloud may offer complete solutions to build, launch, and monitor agents more easily.

Today, the real barrier goes beyond just technology. It is also how people think about agents. Some overestimate what they can do; others are hesitant to try them. The truth lies in the middle. Agents are strong with goal-based and repeatable tasks. They are not ready to replace deep human thinking yet.

The value of agents

Still, the direction is clear. In the next two years, agents will become normal in customer support and software development. Writing code, checking it, and merging it will become faster. Agents will handle more of these steps with less need for back-and-forth. As this grows, companies may create new roles to manage agents, needing someone to track how they are used, make sure they follow rules, and measure how much value they bring. This role could be as common as a data officer in the future.

The hype over AI agents is loud, but the real change is quiet. Agents are not taking over the world; they are just taking over tasks. And in doing that, they are changing how work feels—slowly but surely.

Aravind Chandramouli is vice president, AI Center of Excellence, at Tredence.

—

Generative AI Insights provides a venue for technology leaders to explore and discuss the challenges and opportunities of generative artificial intelligence. The selection is wide-ranging, from technology deep dives to case studies to expert opinion, but also subjective, based on our judgment of which topics and treatments will best serve InfoWorld’s technically sophisticated audience. InfoWorld does not accept marketing collateral for publication and reserves the right to edit all contributed content. Contact doug_dineley@foundryco.com.