https://www.techtarget.com/searchenterpriseai/news/366623681/IBM-customers-assess-the-performance-of-AI-agents
BOSTON -- With agentic AI the tech catchphrase of the year, AI vendors such as IBM are promising customers a chance to build agents and for those agents to be able to talk to one another.
During its Think 2025 conference on Tuesday, IBM introduced new tools in its Watsonx Orchestrate platform. The platform lets enterprises build agents quickly and integrate them with applications from many third-party vendors.
The platform also enables multi-agent orchestration, according to IBM. The approach is similar to Google's Agent2Agent protocol and reinforces the idea of agents moving and communicating among applications and tools.
"AI has moved from experimentation to a real focus on unlocking the business value," IBM CEO Arvind Krishna said during the conference keynote. "Success is going to be defined by integration and business outcomes. Agents are going to help redefine how our people build applications."
The idea of AI agents communicating and working together is innovative, but also complex. Some think the technology is too new and needs more testing and proof-of-concept work.
"My worries are that the AI [technology] is immature," said Armando Castro, director of planning and analytics at Grupo Dportenis, a Mexico-based apparel and accessories company.
While speaking to Informa TechTarget on the first day of the conference, on the expo floor near IBM's Watsonx Orchestrate display, Castro said that while enterprises have expectations about AI agents, the technology needs to develop more before it can lead to actual productivity gains and be safe to use.
"It is a big opportunity for businesses," Castro said. He added that Grupo Dportenis, an IBM customer, has used IBM Watson Studio and experimented with agents, finding them effective. However, despite his company's early success with agents, other companies in Mexico have not had the same positive results.
"We need to understand why all the use cases are not profitable," he said. "We need to know how to work through it."
It's best for enterprises exploring agentic AI technology to watch closely, but be cautious about investing, Castro said.
Financial services company USAA is also taking a guarded approach to agentic AI technology.
"We are moving at speed, but cautiously," said Ramnik Bajaj, senior vice president and chief data analytics and AI officer at USAA, in an interview. "Being a conservative company, we want to manage our risk around these things quite well."
USAA uses IBM's Watsonx Orchestrate platform to automate data management functions. For instance, if a team member needs a data set, an orchestration agent takes the request and decides whether to approve it if it is low risk or send it to the right people if it is higher risk, Bajaj said.
"There's some amount of reasoning, and there's some amount of orchestration workflow," he said. "We are using an orchestration tool to create that workflow and automate those types of back-office processes."
According to Bajaj, the agentic framework is mature enough to use in lower-risk applications.
"It's important for enterprises, our own included, to adopt the agent framework," Bajaj said. He added that large language models on their own, without agents, often are unproductive. However, when LLMs are connected to the tools and trained to take action, enterprises can achieve real automation.
"Adopting agentic AI will be important to achieve that kind of end-to-end automation," he said.
Edward Calvesbert, vice president of product management for Watsonx, said the tools that the agent has access to are key to the success of agentic AI and orchestration.
"Agents need better tools," Calvesbert said in an interview. "The best tool is data." He added that with its Watsonx.data system, IBM is looking to make data accessible and accurate.
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.
06 May 2025