AI on IBM i FAQ

It covers several distinct things that get lumped together: IBM's watsonx product family (watsonx.ai, watsonx Assistant, watsonx Orchestrate), IBM Bob (IBM's GA AI coding agent for IBM i), the IBM i MCP Server (an open-source, beta AI agent connectivity layer for Db2 for i data), on-chip Matrix Math Acceleration in Power10/Power11 hardware, and AI-assisted RPG and COBOL modernization tooling from third-party vendors. See our AI on IBM i overview for how these pieces fit together.

IBM Bob (originally shown as "Project Bob") is IBM's generally available AI coding agent for IBM i, built as a fork of Visual Studio Code and tuned for RPG, COBOL, CL, DDS and SQL. It reached GA 1.0 on March 24, 2026, then Bob 2.0 and the IBM i-specific Bob Premium Package for i on June 24, 2026. Bob 2.0 brought a 270,000-token context window, parallel tool calling, subagents, rollback and three modes; the Premium Package is what lets it read and write source directly in QSYS and the IFS, run CL and SQL, launch 5250 sessions and work on Db2 for i in Database Mode. It can explain code, generate RPG from plain-language prompts, convert fixed-format to free-format, produce documentation and write unit tests. Since August 2026 it can also be launched from watsonx Orchestrate to build agents, currently as a preview. Reported pricing runs from 20 to 240 dollars a month.

watsonx Code Assistant for i was IBM's November 2025 preview announcement of an AI code assistant purpose-built for RPG on IBM i. That preview track never reached independent general availability; it was superseded by IBM Bob, which reached GA on March 24, 2026 and is IBM's current AI coding agent for IBM i. The current product to evaluate is Bob 2.0 with the Bob Premium Package for i, both generally available since June 24, 2026. See our IBM Bob explainer.

Fixed-to-free-format conversion is one of IBM Bob's modernization capabilities, delivered today through the Bob Premium Package for i, generally available since June 24, 2026. Treat it as developer-assisted conversion rather than a fully automatic, unattended process. Confirm current supported RPG constructs and expected developer review effort directly with IBM before scoping a large-scale conversion project around it. This capability was originally previewed under the watsonx Code Assistant for i name, which never shipped independently.

It is an open-source, IBM-published server implementing the Model Context Protocol (MCP), letting AI agents and LLMs such as Claude, GPT-4, and local LLMs query Db2 for i data, inspect system objects, and take defined actions on IBM i using natural language, over a Mapepire connection. It is in beta since October 2025. See our IBM i MCP Server explainer for how it differs from a machine learning feature.

No. The IBM i MCP Server is an AI agent connectivity layer ... it lets an AI agent or LLM query and act on Db2 for i data using natural language over the Model Context Protocol. It does not train or score predictive models, and it is not a substitute for a dedicated machine learning or forecasting capability.

Matrix Math Acceleration (MMA) is an on-chip feature in IBM Power10 and Power11 processors that speeds up the matrix math operations AI inferencing depends on, without a separate GPU. You do not strictly need it to use cloud-based watsonx products, but it matters if you want to run AI inferencing workloads on the same physical server already running your IBM i partitions.

No, not for cloud-based watsonx products like watsonx.ai, watsonx Assistant, or watsonx Orchestrate, or for the downloadable IBM Bob coding agent, all of which run regardless of your underlying hardware generation. Power11 (and Power10) hardware matters specifically if you want on-chip Matrix Math Acceleration for local AI inferencing workloads running alongside IBM i.

General-purpose AI assistants can help with RPG to some degree since RPG appears in their broad training data, but none of them are trained specifically on RPG the way IBM Bob is, and none integrate directly with the IBM i development environment. Treat them as a general coding aid a developer might use ad hoc, not an IBM i-specific or fixed-to-free-format conversion tool. See our AI code assistants comparison for a fuller breakdown.

Start with a narrow, low-risk pilot rather than a platform-wide rollout: pick one well-understood use case (a modernization pilot with IBM Bob, or an agent-querying pilot with the IBM i MCP Server against clean existing data), confirm data and staff readiness first, and expand only after the pilot proves out. See our AI adoption roadmap and readiness checklist for the full planning sequence.

No. watsonx Assistant is a general-purpose conversational AI platform, not built with an IBM i or Db2 for i connector. Connecting it to Db2 for i data to answer questions or trigger IBM i transactions requires custom integration work, the same as connecting it to any other backend system. The IBM i MCP Server is a more direct, purpose-built option for AI-agent access to Db2 for i data, though it is a separate open-source project, not a watsonx feature.

watsonx.ai is IBM's general-purpose foundation-model studio, for building, tuning, governing and deploying AI models; version 2.4 shipped in June 2026 with nothing in it aimed at IBM i. IBM Bob is a separate, IBM i-specific product outside the watsonx brand: a downloadable AI coding agent, built as a VS Code fork, tuned for RPG, COBOL, CL, DDS and SQL, and paired with the Bob Premium Package for i that works directly against QSYS, the IFS and Db2 for i. Use Bob for RPG and COBOL work. Use watsonx.ai if you are building a custom AI application from scratch, and budget for the integration, because it has no Db2 for i connector. The two brands touch in exactly one place today: since August 2026, watsonx Orchestrate can launch Bob to build agents, listed by IBM as a preview.

No. The IBM i MCP Server in particular is accessible to smaller shops because it works through natural language against Db2 for i data you already have, with no separate data science platform required. AI-assisted RPG modernization tools like IBM Bob are relevant to any shop with a meaningful body of fixed-format RPG, regardless of size. The bigger factor is data and staff readiness, not headcount ... see our readiness checklist.

Shipped alongside Bob 2.0 on June 24, 2026, the Premium Package for i is the IBM i-specific add-on to IBM Bob: about 40 skills and workflows and about 30 tools that read and write QSYS and the IFS, run CL commands and SQL, launch a 5250 session, manage library lists and support a retrieval-augmented generation workflow, plus a Developer Mode covering RPG, COBOL, DDS, SQL and CL, and a Db2 for i Database Mode for SQL optimization with error checking. It is what lets Bob work with source code where it already lives instead of requiring a download to a workstation first. IBM's July 15, 2026 Power announcement cites Heartland Co-Op estimating 60 percent faster time for developers new to the platform to understand complex applications with it. Reported pricing at GA was a 20 to 40 dollar monthly add-on on top of a base Bob subscription.

IBM publishes Bob's plans directly. Pro is 20 dollars a month with 50 Bobcoins, Pro+ is 60 dollars with 180, and Ultra is 200 dollars with 1,000. There is a free 30-day trial with 50 Bobcoins, and Enterprise is quoted by sales. Premium packages are billed on top: the IBM i package is 40 dollars a month, so a seat with IBM i support included lands at 60, 100, or 240 dollars. IT Jungle's July 2026 range of 60 to 80 dollars for Pro with Premium reflected an add-on IBM had not yet priced publicly; the IBM i package is now listed at 40.

Bobcoins are the real variable. They are IBM's consumption-based billing metric, spent by each action Bob takes, and once the monthly allowance is gone, enabling overages bills you at 50 cents per Bobcoin. IBM does not publish a rollover policy. It does state that Pro, Pro+, and Ultra are for individual use only, so budget per developer. Prices are indicative and vary by country. Full breakdown: https://pricing.as400software.com/ibm-bob-pricing/

Build with Bob is an option added to watsonx Orchestrate's Create an agent dialog in the August 2026 release. Choosing it launches IBM Bob into a development workspace connected to your Orchestrate environment, so you can generate agents, tools and workflows from plain-language instructions, or refine agents that already exist, instead of hand-writing them. IBM's release notes list it as a preview on IBM Cloud and AWS deployments rather than generally available; the Orchestrate extension on the Bob side is GA. It is the first official link between Bob and the watsonx brand. It does not by itself connect an agent to IBM i data, and nothing in the release notes or announcement mentions IBM i, Db2 for i or RPG. That still needs a tool such as the open-source IBM i MCP Server.

No. Db2 for i has no native vector store as of September 2026, unlike Db2 for Linux, UNIX and Windows, which added a VECTOR data type in its 12.1 release. IBM's stated direction is to run OpenSearch on Power as the vector and hybrid search layer for retrieval-augmented generation against IBM i data, with Db2 for i remaining the system of record. See our RAG on IBM i explainer for what that means for a design.

IBM Power Autonomous Operations is an AI agent for the Power infrastructure underneath IBM i, not for IBM i application code. It continuously monitors Power systems, ingests alerts, runs AI-driven diagnostic analysis, and produces a recommended remedial action that a person approves before it is applied. You drive it with chat-style prompts rather than command syntax. IBM announced it on July 15, 2026 and set general availability for September 23, 2026. The one use case IBM has measured is capacity constraints: 3.33 minutes to detect and resolve a capacity-related condition across eleven Power systems, against 52.59 minutes by hand, which is where the up to 15x claim comes from. It is a different product from IBM Bob, which works on code, and IBM has not yet published licensing, integration points such as the HMC or PowerVC, or IBM i-specific behavior.

In a narrow, real way, yes. The shipped pattern is AI reading scanned bills of lading and populating a freight management system automatically, the kind of feature Carrier Logistics added to its FACTS platform. The dispatch and rating logic itself, on platforms like Trimble's TL2000 or Varsity Logistics' ShipSoft, is not AI-driven; where AI helps there is IBM Bob explaining and modernizing the underlying RPG and CL so a team can safely maintain it. See our full explainer for detail.

Yes to the first, no to the second. Since the August 2026 Orchestrate release, a Build with Bob option in the Create an agent dialog launches Bob into a workspace wired to your Orchestrate tenant, where it generates the agent definition, its tools and the workflow from natural language. IBM lists that option as a preview on IBM Cloud and AWS, not as generally available. What it connects is two build tools, not an agent and your data. Nothing in the release notes or the announcement mentions IBM i, Db2 for i or RPG. To get an Orchestrate agent to Db2 for i you still need a tool that talks to IBM i, and the shortest one available is the open-source IBM i MCP Server, itself still in beta. That three-piece chain works on paper and every part of it is published, but IBM has not documented the combination end to end.

Outside Db2 for i, on a second system. Db2 for i has no vector store as of September 2026, so there is nowhere in the database to put them. IBM's stated direction, from a May 2026 strategy session, is OpenSearch running on Power as the hybrid vector, text and structured search layer, with Db2 for i staying the system of record and OpenSearch holding the embedded, indexed copy an agent searches. Note that IBM said on Power, not on IBM i: in practice that means a Linux on Power partition, and IBM has published no supported IBM i build, reference architecture or timeline. watsonx.ai supplies the embedding models and the text embeddings API that create the vectors, and IBM's own documentation is clear that storing them is your job, in a vector database you provide. See our RAG on IBM i explainer for the full loop and the order to build it in.

Yes, and since July 24, 2026 there is an entry-level machine built for it: the Power S1112, which runs AI inference locally, delivers 2x the core performance of the Power S914 and up to 69 percent better energy efficiency. That puts on-box inference inside reach of shops that were never going to buy a large system for it. The hardware side of that decision lives on our sibling sites: see https://power11.as400system.com/ for the systems themselves and https://ai.as400system.com/ for AI on the platform.

Yes, in four specific ways, and only one of them ships inside a freight product today. AI document extraction reads a scanned bill of lading and creates or updates the shipment automatically, which is the shape Carrier Logistics shipped for its FACTS platform. Fleet maintenance history is strong training data for predicting component failures, though no vendor has published a benchmark for an IBM i edition. The IBM i MCP Server lets an agent answer operations questions off live Db2 for i dispatch and load tables. IBM Bob explains the RPG and CL behind a rating engine nobody has documented in twenty years. AI dispatch and AI route optimization inside an IBM i-native product have not shipped. Settle three governance questions before any of it: what data leaves the box, who approves a write, and what gets journaled.