IBM Corp. is using its Think 2025 conference in Boston this week to introduce a sweeping set of artificial intelligence and hybrid cloud technologies aimed at accelerating enterprise adoption of AI agents.
The venerable tech giant said it’s focused on making the new breed of autonomous software that can take actions independently while working toward a specified goal, easier to build and orchestrate.
The new technologies are designed to address what Chief Executive Arvind Krishna called the “fragmentation of infrastructure” that has stalled many AI initiatives. He cited internal research that found that only 25% of enterprise AI projects have delivered expected returns on investment. IBM is tackling this shortfall with tools for building and orchestrating AI agents, unifying enterprise data and scaling AI inference, the process of running trained models.
Agents in minutes
Central to IBM’s announcements is watsonx Orchestrate, a platform the company claims enables businesses to build AI agents in less than five minutes. The system includes over 150 prebuilt agents, integrations with more than 80 enterprise applications, and orchestration tools that enable multiple agents to collaborate across complex workflows.
“We think this will appeal to both low-code and pro-code developers,” Krishna said in a press briefing. “Agents are going to form an integral part of how generative AI is used in enterprises.”
He said the rise of AI agents will trigger a shift in management thinking. “This is a new way to work,” he said. “C-suite leaders are already using these tools to streamline operations and make decisions. We expect agents to become the interface layer for enterprise users.”
IBM is also launching an Agent Catalog within watsonx Orchestrate to support the discovery and reuse of existing agents. It allows organizations to build domain-specific agents for functions like human resources, sales and procurement and coordinate them internally as well as interact with external applications from vendors like Salesforce Inc., Oracle Corp. and ServiceNow Inc.
IBM Senior Vice President of Software Rob Thomas said IBM is addressing a “big gap in the market for connecting agents.” It has the tools in place, said Brent Ellis, Forrester Research Inc.’s principal analyst serving technology, architecture and delivery professionals, “but the art will be putting those parts together. That will be the hard work of taking things like MCP [Model Context Protocol] and A2A [Agent2Agent Protocol] and generalizing them, as well as adding in key enterprise elements like the ability to pass encrypted data in the model context, or use session keys across multiple platforms,” he said. “Right now, they are doing this in a bespoke way in their consulting organization.”
Hybrid cloud automation
To address the complexity of enterprise landscapes, IBM introduced webMethods Hybrid Integration, a platform that automates the integration of applications, application program interfaces, events and data across on-premises and multicloud environments. IBM said it can replace rigid workflows with intelligent automation powered by agents.
It cited a study conducted by Forrester Consulting that found that organizations using webMethods achieved a 176% return on investment over three years, a 40% reduction in downtime and up to a 67% time savings on routine projects.
“Integration remains one of the most difficult problems in enterprise IT,” Krishna noted. “By using agent-driven integration, we can dramatically cut the time and complexity of those efforts.”
Mining unstructured data
IBM is also revamping watsonx.data, a governed data store designed to scale AI and analytics workloads, with new features that help enterprises extract value from unstructured data such as documents, spreadsheets, presentations and multimedia. Enhancements include open data lakehouse support, data fabric capabilities, lineage tracking and AI-powered tools for orchestrating and analyzing data.
Krishna said connecting AI applications and agents with unstructured data can improve the accuracy of AI applications by up to 40% over conventional retrieval-augmented generation or RAG methods. IBM also introduced content-aware storage for ongoing contextual processing of unstructured data to support real-time inference.
All these capabilities are undergirded by technology IBM recently acquired with DataStax Inc., whose database-as-a-service offering based on the open-source Apache Cassandra database management system is built to handle vast amounts of information distributed across multiple locations and environments. IBM said DataStax brings vector search and real-time data capabilities to its stack.
Mainframe boost
With the introduction of a new generation of mainframes imminent, IBM highlighted the launch of LinuxONE 5, a high-performance Linux platform that it said can process up to 450 billion AI inference operations per day. Powered by the company’s Telum II AI processor and the forthcoming Spyre Accelerator, LinuxONE 5 is aimed at compute-intensive applications like fraud detection and medical imaging.
The platform also introduces confidential containers for secure AI operations and encryption technologies that can’t be easily compromised by quantum processing. IBM claimed a mainframe running LinuxONE 5 delivers a 44% cost savings over five years compared with x86-based systems running equivalent workloads.
Krishna said the announcements collectively reflect IBM’s belief that AI must be embedded into the core of enterprise operations as businesses move beyond experimentation to focus on measurable outcome. “The era of AI experimentation is over,” he said. “Enterprises want integration, governance and real ROI.”
Ritika Gunnar, IBM’s general manager for data and AI, said agents are critical to the transformation. “We are moving from AI that generates content or chats with users to AI that acts on behalf of users,” she said. “These agents must integrate with existing systems, workflows and governance requirements.” She said IBM expects more than 50% of enterprises will embed agents in essential systems by 2027.
Governing agents
Asked what it plans to do about AI hallucinations and bias, IBM said its governance tools have been extended to support agents and that the company is investing in lifecycle management, observability and traceability for agentic workflows. “Just like we evaluate and govern large language models, we’re doing the same for agents,” Gunnar said.
IBM supports open standards such as MCP developed by Anthropic PBC and has open-sourced its Agent Collaboration Protocol to promote interoperability and agent-to-agent communication. The company has also open-sourced its Granite model family and continues to partner with open model developers. “We think the future of AI is open,” Thomas said.
Forrester’s Ellis said that though few companies outside of the IBM customer base have adopted watsonx or Granite at this point, the company’s commitment to openness should resonate with buyers in the long term.
“IBM’s revenue model is on the services and infrastructure used to run these models, and it would make sense for them to keep their models open,” he said. “The need for models that address governance concerns like security and indemnity will grow, as will the need to ingest historical data for use within AI models. IBM can address both of those trends. This looks good for IBM.”
Image: News/Microsoft Designer
Your vote of support is important to us and it helps us keep the content FREE.
One click below supports our mission to provide free, deep, and relevant content.
Join our community on YouTube
Join the community that includes more than 15,000 #CubeAlumni experts, including Amazon.com CEO Andy Jassy, Dell Technologies founder and CEO Michael Dell, Intel CEO Pat Gelsinger, and many more luminaries and experts.
THANK YOU
