AI Is Coming for ERP. How Will IBM i Respond?
September 8, 2025 Alex Woodie
A wave of artificial intelligence is beginning to wash over the enterprise resource planning software market, changing the nature of how businesses process automation gets done. IBM i has been a solid platform for running ERP systems for decades. But how will the new generation of AI-enabled enterprise apps get to IBM i?
When it comes to automating business processes, ERP packages have been the primary means for achieving it for the past four decades. The big monolithic apps from SAP, Oracle, Infor, and dozens of others have driven trillions of dollars in value by providing a set of common business rules that standardize various business functions and also unifying the data that companies use to power sales, marketing, billing, manufacturing, inventory, logistics, support, reporting, etc. In the IBM i market, a large number of companies have also created their own custom ERP packages, or cobbled a mix of homegrown and third party applications together to encapsulate their business processes.
Nonetheless, big packaged apps have also made their way into the IBM i world and they often set the pace of innovation for the homegrown crowd. SAP, Oracle, and Infor continue to support mainstream, tier-one ERP applications running natively on IBM i (or through PASE), although both SAP and Oracle have set deadlines for customers to move off IBM i platforms to continue to receive support. However, about 70 percent of IBM i shops develop their own applications, according to the latest IBM i Marketplace Study from Fortra. That is arguably a higher rate of custom development than one sees on other platforms, and it colors the future of IBM i and the business applications that run on it.
The broader enterprise computing market – which includes ERP and related functions (CRM, SCM, MRP, etc.) – has seen some big shifts over the past couple of decades. The advent of cloud and mobile computing in the 2007-2015 timeframe uncoupled ERP from in-house datacenters and Windows desktops, freeing customers to participate in SaaS-enabled ERP workflows from the comfort of their mobile devices. Instead of big monolithic code bases running on giant symmetric multiprocessor (SMP) servers, the state-of-the art shifted toward running cloud-based containerized microservices that companies could update frequently and run independently without fear of breaking the whole ERP system.
In the midst of this massive change, we got another massive change in the form of generative AI. Since ChatGPT debuted in late 2022, the entire business community has been trying to figure out how to best take advantage of large language models (LLMs) that display an uncanny capability to understand and generate language. This has unlocked enormous potential across the global business community. In fact, in 2023 McKinsey said GenAI could drive anywhere from $2.6 trillion to $4.4 trillion in annual economic gains.
2025 IBM i Marketplace Study by Fortra.
Chatbots were the first GenAI vehicle for leverage LLMs. Instead of paying with a live human customer service agent to interact with customers, companies substituted LLM-powered chatbots to handle common requests. Some enterprising companies developed their own chatbots based on open source LLMs, such as Meta’s Llama-2, but most just used cloud-based LLMs developed by tech giants like OpenAI, Google, and Anthropic.
In some cases, companies built elaborate retrieval augmented retrieval (RAG) workflows that involved indexing data from their own internal knowledge bases into vector databases, which they could then combine with content from customer requests and feed into the top foundation models running in the cloud.
Coding copilots also provided a productivity boost to developers. LLMs can understand programming languages and generate understandable code, such as Java, PHP, or SQL, in the same way they can understand English or Mandarin. IBM Rochester is currently developing a copilot that’s trained on RPG. Watson Code Assist, as the product is called, is currently in beta, and is due to be ready for prime time later this year.
If chatbots and copilots represent the first generation of generative AI tools in enterprise computing environments, then agentic AI will represent the next generation. Lots of the excitement around agentic AI is based in large part on the emergence of so-called “reasoning models” such as DeepSeek R1 and OpenAI o3 late last year and in early 2025.
Reasoning models differ from LLMs such as OpenAI’s GPT-4 in how they work. Instead of cramming the “context window” of an AI model with oodles of input, a reasoning model works by breaking down a request into separate components, and then working on the individual pieces separately, the so-called “chain of thought” process that’s somewhat analogous to how humans think.
WCA for i is built on watsonx and Granite models.
Much of the current interest in AI is centered on the potential of agentic AI to fundamentally transform how work gets done. If a business process currently driven or performed by a human can be codified in language, then it theoretically should be automatable through reasoning models and agentic AI.
The pace of innovation in the agentic world is accelerating. It was only ten months ago that Anthropic suggested Model Context Protocol (MCP) as a potential way connecting AI agents with data sources. Over the past few months, MCP adoption has skyrocketed, and today most databases and AI tools support MCP as a sort of API for enabling AI workflows, similar to how REST provides the underpinning for modern microservices.
The impact of agentic AI will be felt across the ERP landscape. Currently, only 5 percent enterprise applications have task-specific AI agents built into them, Gartner said last month. However, by the end of 2026, that figure will skyrocket to 40 percent. We’re currently at an inflection point in the future of AI-enabled enterprise apps, Gartner said, and software companies have only a three-to-six-month window to define their AI product strategy.
(Gartner’s predictions are also changing quite rapidly. In June, Gartner said that 33 percent of enterprise apps will have agentic AI by 2028. The timetable appears to be compressing, according to Gartner.)
AI assistants and task-specific agents represent the bottom floor for AI enablement in the enterprise. Gartner foresees a future where AI agents begin collaborating with each other and solving more challenging problems. By 2027, Gartner sees one-third of agentic AI implementations “combining agents with different skills to manage complex tasks within application and data environments,” the analyst group says. By 2028, AI agents will begin working across multiple applications and business functions.
“AI agents will evolve rapidly, progressing from task and application specific agents to agentic ecosystems,” stated Anushree Verma, a Gartner analyst. “As agentic AI matures, standardized protocols and frameworks will enable seamless interoperability, allowing agents to sense their environments, orchestrate projects and support a wide range of business scenarios.”
Gartner predicts agentic AI having a big impact on enterprise applications.
As AI agents begin doing more of the actual work, it will change the role of the remaining humans. Instead of making decisions and powering the workflow in the ERP system, human workers will transition to more of a supervisory role, managing the AI agents. Developers who have AI skills will also be in high demand, as will be folks with experience setting guardrails and governance standards.
How will IBM i shops and IBM i software vendors adapt to this new AI reality? While it’s theoretically possible to retrofit existing ERP systems to adopt these new technologies, it may not be worth it. Today’s ERP systems are built on a deterministic foundation, where languages like RPG are used to create business rules that can be enforced. In future ERP systems, many more decisions will be made by AI algorithms, which are probabilistic systems without set rules. While ERP systems are unlikely to completely abandon rules-based approach for some components (a probabilistic general ledger, anyone?), the winning ERP systems of the future will likely come from a clean-slate approach.
It’s unlikely that ERP companies of the future will target the IBM i with such a clean-slate AI-enabled ERP system, at least for the tier one mainstream ERP systems (the tier two and specialty ERP providers could maintain their IBM i support). That’s not a knock on the platform, but more just a fact of how business is done today. It’s more likely that tier-one mainstream ERP systems will be sold primarily via the SaaS model, as SAP is doing with its current S/4 HANA and other ERP vendors are doing with theirs.
As for the lion’s share of the IBM i installed base that currently roll their own code and will continue to do so, the future would appear to be wide open. The good news is that the IBM i platform itself is fully capable of running modern code developed using modern coding practices, which includes agentic AI. The widely held perception that IBM i is “old” is due mainly to the platform’s industry-leading backwards compatibility and the installed base’s (sometimes questionable) decision to avail themselves to it.
Today’s IBM i developers are already using copilots to help them write better code. The pace of development in the world of open source software – including open source AI software – is tremendous. It would seem that the odds are quite high that some enterprising coder will develop a framework that allows organizations to incorporate modern agentic AI workflows into their existing systems. In fact, it might already exist, and there’s no reason it can’t run on IBM i.
RELATED STORIES
RPG Code Generation And The Agentic Future Of IBM i
Public Preview For Watson Code Assistant for i Available Soon
