IBM i modernization with AI: tools, pricing, where to start

IBM i modernization with AI in 2026 finally has real tools, real pricing, and real numbers to argue about. The reason your shop is suddenly paying attention isn’t trend chasing. It’s that the developers who actually understand your RPG are getting close to retirement, and AI is the first capture tool that has a chance of working before they leave. This is what you should know, what it costs, and where to start this week.

The retirement clock is the real driver

A 2025 industry survey found 73% of IBM i shops now prioritize AI integration and automation. That number is not about chasing GenAI hype. Read the COMMON user group panels and the IT Jungle commentary from this spring and the through-line is the same: experienced RPG and COBOL developers are retiring faster than shops can replace them, and the people who remember why a 30 year old subfile program was written the way it was are not going to be at their desk next year.

AI is interesting on IBM i specifically because it is finally good enough to read RPG IV and explain it back to you in language a junior developer can understand. That is the shift. Everything else, the conversion tools, the IDE assistants, the agent frameworks, is downstream of that one capability.

Three things AI is actually useful for on IBM i right now

Skip the marketing decks. Right now, here is what these tools genuinely do well.

Documentation generation. Point a modern LLM at an undocumented RPG program and ask it to explain the program flow, the file inputs and outputs, the business rules embedded in the calc specs, and the dependencies on other modules. The output is not perfect, but it is a 70% draft a developer can finish in an hour instead of a week. Our guide to documenting RPG IV with AI walks through the prompt templates that produce usable output the first time.

Code understanding for new developers. A two year RPG developer with Claude or Bob in front of them can read a fixed format program written in 1998 and form a working mental model of it in an afternoon. Without AI, that same exercise is a multi day exercise and the developer is guessing for half of it.

Selective conversion. Fixed format to free format RPG. Embedded SQL cleanup. Pulling business rules out of monolithic programs into service programs. AI is solid here as long as a human reviews the output. Full unattended conversion of complex programs is not solved yet, and I will come back to that. Testing the converted SQL is easiest in the tool most shops already have open for it: our IBM i Access Client Solutions review covers how Run SQL Scripts and Visual Explain make it straightforward to confirm a converted statement performs as well as the original.

What AI is not yet useful for: replacing your green screen with a modern UI, generating a complete net new application from a vague spec, or migrating a mission critical RPG program to Java without a human in the loop for weeks of testing. If you are weighing a platform move at all, what actually happens when IBM i shops migrate is worth reading before you commit.

IBM Bob: the native path and what it actually costs

IBM Bob 1.0 went generally available on March 24, 2026, with the global launch announced April 28. It is IBM’s first serious AI development partner that targets IBM i directly, and unlike the watsonx Code Assistant, it is sold to individual developers and teams without a six figure enterprise contract.

The pricing model is the part most IBM i developers get confused by. Bob sells in credits called Bobcoins. The current tiers as of June 2026:

• Free trial: 30 days, 40 Bobcoins included.
• Pro: $20 per user per month plus $3 per month support, includes 40 Bobcoins.
• Pro Plus: $60 per user per month plus $9 per month support, includes 160 Bobcoins.
• Ultra: $200 per user per month plus $30 per month support, includes 500 Bobcoins.
• Enterprise: approximately $500 per month per resource unit with SSO and quota management, sold as a separate SKU.

A Bobcoin is consumed per task, with cost varying by the size and type of the work. In practice, a developer doing four or five code explanation prompts and a handful of conversion tasks in a workday will burn through the Pro plan’s 40 Bobcoins quickly. For active daily use on real IBM i work, Pro Plus is the realistic starting tier.

Bob supports RPG, CL, SQL, COBOL, Java, and Python on IBM i. The big news is the Bob Premium Package for i, announced on May 21 and releasing on June 24, 2026. It adds an IBM i Developer Mode tuned for RPG, CL, and SQL, roughly 40 new skills including fixed format to free format conversion and business rule extraction, RAG augmented with IBM i specific knowledge, and direct access to QSYS source members. If you are running the free trial right now, schedule the next evaluation for late June when the Premium Package lands. The pre-Premium Bob is good. The Premium-augmented Bob is what you actually want to benchmark.

IBM’s own reported numbers from the launch: an average 45% productivity gain across users, and the Instana team reporting a 70% time reduction on certain workflows. Take vendor self-reported numbers with the usual skepticism, but the underlying capability is real. IBM shipped a major architecture update in July, covered in what actually changed for IBM i shops with Bob 2.0, if you want the current state of the platform rather than the March launch numbers.

The third-party toolkit: Fresche, ARCAD, and Remain

Three non-IBM tools matter, and our broader survey of AI tools for IBM i developers covers each in more depth. Each solves a different problem.

Fresche X-Modernize AI. Launched November 2025 as a third generation RPG-to-Java and RPG-to-PHP conversion tool. Fresche claims 50% faster automated conversion with preserved business logic. The architecture is a multi agent, LLM agnostic framework sold as Modernization-as-a-Service, meaning monthly subscription with no upfront license, custom quoted per engagement. There is no public price list, and our full review covers what that free evaluation actually gets you, plus how much weight to put on the 50% claim. The right move is to take the free no obligation application evaluation, which gives you a concrete sizing and price for your specific code base. If you are seriously considering a multi year RPG-to-Java conversion program, Fresche is the one to talk to. If you are not, do not pretend you are.

ARCAD MCP Server. This is the most interesting product for shops that already have Claude, Copilot, or Bob, and want to plug their IBM i application context into those agents. ARCAD MCP Server exposes 70 plus MCP tools that give an AI agent structured, validated context from the ARCAD repository, including impact analysis, cross reference diagrams, and code context. The pitch is hallucination reduction: instead of asking Claude to reason about your program from raw source, you give it a curated and accurate view of the program’s dependencies. New MCP tool batches drop every three months. By spring 2026, coverage included DevOps, DROPS deployment, DISCOVER analysis, and the free format converter. Pricing is not public. If you already pay for ARCAD, this is a clean upgrade. If you do not, the entry cost is significant. Giving an agent that much reach into your system is worth pairing with the same authority discipline you would apply to any user profile, our guide to what to lock down before connecting AI agents to IBM i covers the specifics.

Remain Software AI Chat. First to market in 2023 and the easiest entry point. It integrates with RDi and MiWorkplace, uses ChatGPT under the hood, and is available as freeware on GitHub. If you want a working IBM i AI tool in your IDE today, with no procurement conversation and no Bobcoin math, this is the answer. Paid versions exist for organizational use.

There is a fourth option worth mentioning briefly: IBM watsonx Code Assistant for i, built on the Granite model and tuned for RPG. It has been in preview since 2025 and the public pricing for the IBM i variant is not clearly published. For most shops in 2026, Bob is the cleaner story and the Granite tooling can wait.

Where to start: a practical sequence

Stop reading vendor PDFs and do this.

Week 1. Pick one undocumented RPG program. Something nobody wants to touch, written by someone who left, that you have been meaning to understand for two years. Sign up for the Bob free trial. Burn 10 Bobcoins asking it to explain the program flow, list the file dependencies, and extract the business rules. Read the output critically. This single exercise tells you more about what AI can do on IBM i than any vendor demo.

Weeks 2 to 4. Build a documentation prompt that produces consistent output across multiple programs. This is harder than it sounds because RPG style varies wildly across shops and decades. Once you have a prompt that produces a usable artifact every time, you have a tool. Run it on the next five programs in your backlog.

Month 2. Assess. If the value you are getting is documentation and code explanation, stay on Bob Pro Plus and add ARCAD MCP if you have the ARCAD repository already. If you have a board mandate to convert RPG to Java in the next 24 months, get on a call with Fresche and stop pretending you can DIY it. If you just want a free baseline for the team to play with, drop Remain AI Chat into RDi this week. And if this is also the moment you’re evaluating whether to move off RDi entirely, the 2026 IDE comparison is worth reading first.

The mistake every IBM i shop makes is starting with a six month evaluation committee. The right move is to spend $20 on a Bob trial and find out in week one whether this works for your code base.

What AI can’t do yet on IBM i

Be honest. Three things are still not solved.

Embedded business logic in monolithic RPG programs is hit or miss to extract cleanly. AI will pull out the obvious rules and miss the subtle ones. A human has to review.

Green screen replacement is a UI problem, not a code generation problem, and no current AI tool ships you a working modern interface. The conversion tools convert logic, not UX.

Full unattended RPG-to-Java conversion is still not real. Fresche’s 50% faster claim is 50% faster than the previous generation of tools, not 50% faster than a human team doing the whole job. You still need engineers running QA, building the test harness, and fixing the 10% the converter gets wrong.

If a vendor tells you their tool does any of the above with no human in the loop, get the demo data and prove it on your code, not theirs.

The knowledge transfer play

The honest reason to move on this now is the demographic one. Bob can sit down with a module written by someone who retired in 2018 and explain it in language a developer hired last quarter can follow. That is a knowledge transfer tool dressed up as a coding assistant. If you have senior RPG developers within a few years of retirement, the right project this quarter is not a Java conversion. It is using AI to capture, document, and explain every program they own while they are still at the desk to verify the output.

If you want a structured playbook for running that capture project, including the 35 IBM i specific prompts and workflow we put together for shops doing exactly this work, the IBM i AI Field Guide is built for it.

Bottom line

Buy a Bob Pro Plus seat for one developer this week, give them an undocumented RPG program, and see what happens. That single exercise will tell you more than three months of vendor evaluations. Wait until late June for the Premium Package for i to land before scaling to the rest of the team. If you have a board level mandate for full RPG-to-Java conversion as part of an IBM i modernization roadmap, call Fresche for the free assessment. If you do not, focus on documentation, code understanding, and knowledge transfer while the people who remember why the code looks the way it does are still on payroll. That window is the one that closes, and it is the only one that matters.