The Big Easy: Connecting The Dots On Big Blue’s AI Strategy For IBM i

May 4, 2026 Alex Woodie

IBM has not yet formally announced the new AI products that it is building for IBM i. We know about Bob and the MCP server, but there is a lot more AI stuff coming to IBM i. IBM shared details of what is coming in closed sessions at POWERUp in New Orleans last week, but hints about where it is going and what is coming were there, if you knew where to look.

IBM i, as we all know, is a business platform. It runs enterprise applications and it runs them extremely well. If you want to model a nuclear reaction or train an AI model, it is not the right system. However, if you want to run an AI model with production enterprise applications, you would be hard pressed to find a more perfect system than IBM i

That, in a nutshell, is the pitch that IBM made at COMMON’s annual conference last week. During her keynote address during the Opening Session, Power Systems general manager and IBM Fellow Hillery Hunter drilled home the assertion that the combination of AI and IBM i is destined to be a fruitful one.

Customers running on open systems have big decisions to make in terms of how they construct a data lake to support their AI endeavors, Hunter said during her keynote. But IBM i shops are largely spared from those decisions, since it already provides the strong data foundation that is lacking in open systems.

“There are very complex decisions where you have to work through the data lake, the data warehouse, the data governance conversation, where you have to acquire new middleware to tackle those problems so that you can enter into the AI era,” Hunter said. “We have all those products for the large estates that need to be going through those explorations. But right now, if you’re on IBM i and your workflow is running there…you can get going with that foundation that you have and fully enter into this AI era.”

AI On Power

During his session “AI with IBM Power – portfolio and strategy,” Ashwin Srinivas, senior product manager AI for IBM Power, provided a glimpse into how IBM is thinking about the challenges and opportunities that AI present to IBM i.

IBM wants to keep AI resident on Power, if not all in IBM i.

“AI native is sort of becoming the new default in enterprises,” he said. “Business units are looking for AI native workflows where . . . a lot of business-critical workflows actually execute on the platform.”

Bob is the AI product for software development on IBM i. IBM launched the first version of Bob in March, and it will soon start providing a premium pack that provides important capabilities, including support for reading source code files directly from IBM i instead of being forced to download source to your PC or read it from the cloud.

“A lot of enterprise architects are looking at how code can be modified, built faster, modernized, assisted in explaining or all that kind of stuff with AI native development,” Srinivas said. “Last but not least, we are looking at AI-native operations: How to keep the platform constantly updated, managed, patched, secure, so on and so forth, with AI first.”

Srinivasan said IBM is working to connect the computing resources to power AI with the data that makes it all work for an enterprise. While all of the pieces will reside within a Power box, it may not all reside under IBM i.

“We try and run AI as close to the data as possible. So all the critical data lives on the Power platform,” he said. “We are building a platform that includes the necessary infrastructure, the data layer, and a few other capabilities that allows you to run AI so that it is integrated with enterprise processes on the same platform.”

That goal may involve running some AI processes within OpenShift, which is Red Hat’s Kubernetes implementation for its Enterprise Linux distro. IBM is not interested in making Power a general-purpose platform for training AI models. The goal is purely to enable AI for enterprise customers. In some cases, that could mean training some smaller open models on Power and its accelerators or (more likely) doing some fine-tuning of AI models based on feedback.

“We’re trying to build out some turnkey AI capabilities with Spyre,” Srinivas said. IBM is also exploring some other AI accelerators beyond Spyre to “unlock further acceleration, larger models, higher concurrency, higher throughput,” he said.

IBM has focused predominantly on running AI and machine learning on structured data, but IBM is looking to expand its capability to process unstructured data using its Watsonx platform, he said. Context is key in AI, and IBM is working to ensure that context is maintained as agents begin to operate on data residing in various Power platforms.

“You actually have to orchestrate across those agents to make sure that your business process or the workflows that you’re building, whether it’s business, enterprise or code, is actually done correctly with each of those agents doing a particular step,” he said. “So we are looking to build out some orchestration platforms across all of IBM i, AIX, Linux, so on and so forth.”

Vector databases play key roles in retrieval augmented generation (RAG) pipelines, as they provide AI applications with extremely quick access to specific pieces of information that have previously been vectorized. However, the Db2 for i database currently does not support a vector store.

“There are multiple ways of doing it,” Srinivas said. “We support a few of those. We don’t support the storage on Db2 yet. So right now we can store in dedicated store like this. But as IBM, what we are trying to do is switch to something called OpenSearch. OpenSearch allows you to search across vector spaces, across text, across structured data, because it’s a hybrid search engine.”

Agentic Native Operating System

IBM engineer Adam Shedivy provided a glimpse of what AI-enabled IBM i operations might look like in his session, dubbed IBM i: The First Agentic Native Operating System.

As one of the young engineers IBM has hired over the past few years, Shedivy normally works on development tools, such as the Code for i plug-in. So he was not aware of the vast assortment of SQL services that IBM’s database architect Scott Forstie and his team have been building into ACS and the Run SQL Scripts facility.

IBM i possess the core features necessary for an agentic AI operating system.

“It’s kind of funny because I didn’t realize that we had all these ways to get information on the system, like SQL services. I never really used them in my day to day,” Shedivy said. “I think we were at a CEAC meeting and Scott was like ‘Oh, I have this SQL script that can detect vulnerable files and it can actually even generate the commands to patch those files.’”

Forstie, of course, has been preaching the gospel of SQL-based automation for years. Human administrators can kick these commands off via ACS and the Run SQL Scripts facility. But the real power of SQL exists through automation, when these scripts can be kicked off in response to other events or linked together to accomplish specific tasks.

In another example of technological fortuitousness, the advent of Model Context Protocol (MCP) has become the industry standard for connecting AI models with existing data sources. The potential for introducing an AI agent to IBM i data and resources is ripe for the taking, for anybody who wants to build it themselves. As Shedivy noted, 30 minutes of Python coding will get you there.

“You’re not strapping on AI assistants or a chatbot window around your data. You’re actually building these agents that now participate in your operating model,” Shedivy said. “These agents can actually reason and decide and act over your business in an effective way. And it turns out this is what everyone is kind of striving to be. Everyone wants to be AI native. Everyone wants their businesses to be 100 percent query-able by AI. Turns out IBM i is actually in a very good position to support a lot of this.”

To work effectively on enterprise IT systems, according to Shedivy, an agentic system must have five core capabilities: scheduling, identity, state management, auditability, and policy enforcement (security). Shedivy mentioned how startups like Runlayer are seeking to build agentic platforms that layer these capabilities on top of enterprise computing resources. Turns out, IBM i’s already got it.

“It turns out that all these cool things like identity bound execution, scoped authority, audibility – those are all things that come with IBM i because of its integrated nature,” he said. “IBM i actually has all five of these out of the box.”

The New New Backend

During his keynote, Agentic AI and IBM i, business architect Jesse Gorzinski immediately addressed the elephant in the room. “Let’s start with a round of applause for the coining of the term ‘Agentic native operating system.’ A round of applause for Adam,” Gorzinski said. “It’s absolutely beautiful. But next week, I’ll take credit for it.”

Here’s another phrase that Gorzinski may not have coined but also may try to take credit for: “Yesterday’s frontends are today’s backends.”

In other words, in the same way that humans drove the 5250 greenscreens or GUIs that hooked into traditional backend systems (such as payroll processing, HR tasks, or invoice generation), AI agents are now being marshalled and driven to get work done by essentially emulating the users tapping the keyboard and moving the mouse to click buttons on the GUI. The original backend (i.e. the IBM i) is assumed to be stable, secure, and available, but the path into it has now moved up a level into the agentic realm.

Gorzinski created an AI agent that automatically checked the IBM i for security vulnerabilities and generated this report.

Gorzinski provided several examples of how AI agents can interact with IBM i systems. For instance, you could use the IBM i-based MCP server to connect an AI model with sales data sitting in Db2 for i. Then you can prompt the model to build a PowerPoint presentation with any interesting insights or trends that the model finds in the data. Gorzinski actually did this, then showed the results to his COMMON crowd.

“That’s no $80,000 software package. That’s a couple of SQL statements, an MCP tool, and a reasoning large language model,” Gorzinski said. “So people like me that build PowerPoint slides for a living have to find something else to do,” he joked.

Gorzinski showed several other examples: The trucking firm that tapped Real Vision Software to build a problem resolution app for truckers; an app that designed and priced out the materials for a flower bed; and IBM’s own Zero Client initiative, which reportedly saved the company $4 billion in human resources costs by replacing humans with AI.

When it comes to AI and IBM i, there is a lot going on behind the scenes. Just as IBM was not quite ready to announce the spring Technology Refresh in time for the annual COMMON conference, the company is not quite ready to publicly disclose everything it’s got cooking in the lab.

“I can’t tell you everything that’s going on. I can’t tell you everything that’s coming,” Gorzinski said. “Maybe if the conference were a couple months later, maybe I could, but I can’t talk about them yet. But what I can do is I can put a couple dots on the paper and hand you a pen and let you connect the dots.”

This Issue Sponsored By

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search