No surprise that AI was the overarching theme at the IBM Think conference this year. "The era of AI experimentation is over,” declared Arvind Krishna in his keynote. “Today's competitive advantage comes from purpose-built AI integration that drives measurable business outcomes," he said.
Among the many talks at the conference (which you can watch on demand here), there was a raft of exciting generative AI-related IBM product announcements - many in preview to provide a glimpse of what’s coming down the line soon. Here’s a snapshot of five launches I’ve read about that I think are particularly interesting.
1. No-code Agent Builder that lets you build an AI agent in under 5 minutes
Among many agentic AI announcements was Agent Builder, a new tool that’s part of the wider IBM® watsonx Orchestrate platform, which allows enterprises to build, deploy and manage agents to automate workflows and processes with generative AI.
Agent Builder, which includes no-code tools that make it possible to build an agent in under 5 minutes, provides a framework that simplifies the process of building, customizing and deploying agents.
There is scope to customize a variety of pre-built agents and tools developed by IBM, as well as by partners such as Salesforce, 11x, and SAP. This includes agents to handle specialist business functions like HR, sales and procurement, as well as utility agents for simpler tasks like web research and making calculations. Agents can be grounded and integrated with a firm’s own content and systems.
Agent Builder is part of the IBM watsonx Orchestrate platform, which provides the ability to combine multiple agents for enterprise-wide collaboration and workflows with appropriate guardrails and controls.
2. Watsonx.ai Model Gateway: the AI-agnostic gateway for accessing multiple top foundation models
IBM’s watsonx.ai Model Gateway gives enterprises the ability to run the best foundation models for their specific needs, regardless of model vendor, where the model runs, or the underlying infrastructure it's hosted on. In short, this is an AI-agnostic gateway that provides the freedom and flexibility to integrate and manage multiple AI models across environments while optimizing costs and ensuring governance.
With the Model Gateway, organizations have a secure, governed, controlled platform to access top foundation models, including IBM Granite, OpenAI, Anthropic, Google, NVIDIA, and more.
“You don’t need to choose one model or platform anymore: you can choose all of them, on your terms,” Armand Ruiz, VP of AI Platform at IBM, said on LinkedIn. “And when the next breakthrough model lands next quarter, or next week, you’re ready. We’re building for a world where the best AI for the job may live anywhere. And watsonx is becoming the AI operating layer to bring it all together.”
3. watsonx Code Assistant for i: AI-powered coding for RPG developers
The watsonx Code Assistant for i is the new AI coding assistant that’s purpose-built for modernization of IBM i applications. Currently available in preview mode, it promises to empower developers who use RPG (the most common IBM i programming language) with AI-powered capabilities accessible through their integrated development environment (IDE).
Thousands of enterprises currently run business-critical applications on IBM i thanks to its reliability, performance and low cost of ownership. But maintaining and modernizing these systems has become increasingly difficult due to the shortage of skilled RPG developers.
Built on the IBM Granite code model and fine-tuned for RPG, the new code assistant is designed to tackle the IBM i skills crisis. It combines the power of Gen AI and advanced automation to simplify coding - increasing productivity and helping developers at all experience levels deliver high-quality code efficiently. It includes the ability to work with fixed-format RPG as well as free-format.
4. IBM Granite 4.0 Tiny Preview: smallest Granite model runs on low-cost consumer hardware
IBM Granite 4.0 Tiny Preview is a preliminary version of the smallest model in the upcoming Granite 4.0 family of language models. Underlining IBM’s commitment to efficiency and practicality in enterprise AI, this compact and compute-efficient model is able to run on inexpensive consumer-grade hardware (including GPUs commonly available for under $350 USD).
While only partially trained with 2.5T of a planned 15T or more training tokens, it delivers performance that rivals that of IBM Granite 3.3 2B Instruct despite having fewer active parameters and roughly 72% lower memory requirements.
The Granite 4.0 Tiny Preview version is now available on Hugging Face with the aim of allowing even GPU-poor developers to experiment and tinker with the model on consumer-grade GPUs. It will be officially released in summer 2025 as part of a model lineup that also includes Granite 4.0 Small and Granite 4.0 Medium.
5 New enhanced watsonx.ai text extraction API
A new text extraction API has been added to watsonx.ai to simplify the process of extracting text from a variety of file formats. The latest version of document understanding technology developed by IBM includes support for a wide range of input and output file formats, including Microsoft Word documents, Microsoft PowerPoint presentations, HTML, and various image formats. It also combines improved image processing and diagram verbalization for accessibility and interpretation with Optical Character Recognition to extract text in machine-print in over 100 languages in multiple scripts and handwriting in English. It can extract structured data in key-value pair format from specific document types such as invoices and utility bills.
These five announcements signal IBM’s clear direction of travel for generative AI - moving it from ‘experimentation mode’ to supporting practical deployment by enterprises.