March 30, 2026

One of the early testers of the IBM Bob AI tool is Medhost, the longtime developer of an integrated IBM i-based healthcare applications for community and rural hospitals. According to a joint presentation between IBM and the software vendor last week, the beta test was a success, as Bob assisted Medhost in not only understanding its RPG and SQL codebases, but modernizing and enhancing them.

IBM i chief architect Steve Will was joined by Michael Bowen, the senior vice president of technology at Medhost, in a webinar last week to discuss how the provider of electronic health records (EHR) software adopted IBM Bob, the new AI-powered coding assistant that coincidentally also became generally available last week.

Bowen had good things to say about the beta version of Bob, which Medhost developers tried out used alongside other AI assistants in 2025 and early 2026. While Medhost didn’t test Bob widely with its developers and engineers, in the limited projects and use cases it has been involved in, it’s been a hit.

Specifically, Medhost has adopted Bob in several areas, including cross-referencing and impact analysis; code generation; converting older version of RPG to free form RPG; and generating better SQL statements.

In the past, cross-referencing and impact analysis was an “extremely time-consuming effort,” Bowen said during the webinar with Will. But Bob accelerated that code analysis and has proved to be “a huge help.”

“The feedback we’ve gotten from our software engineers . . . we’ve had good success with it, providing the engineer the intelligence to understand really a great deal of information quickly about a particular program,” Bowen said. “It’s taking out a lot of that risk of missing program uses and dependencies that a more manual review across many, many different engineers could introduce.”

When it comes to RPG code generation, the early returns on Bob are promising, according to Bowen. “We’ve seen its ability to cut down manual coding efforts pretty significantly, especially compared to no AI code assist at all,” he said. “In some cases, certain projects we’re seeing up to 50 percent reduction in our experimentation, which is hugely impactful.”

Bob shows promise as a tool to help Medhost modernizing its IBM i applications, including moving to free-form RPG and upgrading the database access by generating better SQL, according to Bowen.

“I think a good real-world example that that many organizations face when, when dealing with previous versions of RPG code, would be around internally defined database file structures,” Bowen said. “When you talk about expanding database fields and these structures without the use of an AI tool like IBM Bob, it makes it a very manual process, especially when it has to be completed across a large number of programs.

“By utilizing Bob, what we kind of did through our experimentation process is we were able to convert older versions of RPG to free format, and then convert our data access to SQL and data access objects,” he continued. “And then at that point the manual effort really gets much smaller because . . . we just have to convert the database tables.”

Medhost also used Bob to optimize some of its existing SQL statements. According to Bowen, the engineers and reached the limit on how optimized they could make certain SQL statements that are highly complex. Medhost took those SQL statements and asked Bob to optimize them, and Bob successfully did it, Bowen said.

“It was able to consolidate those multiple SQL statements into a much more efficient single statement, and it made significant optimization from a performance standpoint to those two programs,” Bowen said. “It revealed to us how much IBM i’s SQL capabilities have evolved and how to implement some features that we weren’t aware existed for Db2 on the i.”

Throughout his conversation with Will, Bowen emphasized that Medhost will not skimp on code quality in its pursuit of increasing the pace of development and engineering work. Every piece of AI-generated code goes through the same quality checks as part of its software development lifecycle as human generated code.

“AI is a force multiplier,” Bowen said. “Our goal is we want to create more throughput to produce more efficient efficiency, but continue to generate top quality. And what we found is really meshing the acceleration of our AI tools with our subject matter experts and our technical experts – that’s really the secret sauce. It’s really kind of a collaboration with the AI tool and our resources.”

The quality of AI output is impacted by a variety of factors, Bowen said, chief among them the quality of code documentation and the quality of the prompts given by the user. Bowen said Medhost’s code is fairly well-documented, but as a 40-year-old software company, there are some parts of the codebase that could see some improvements.

“Good comments, good documentation just really increases the efficiency of the process to ensure you get better results and that you’re not going back and forth multiple times and having to continue to provide more context,” he said.

Other variables that factor in AI success include getting buy-in from users. Taking an iterative approach and breaking up a project into smaller chunks is also beneficial with AI-assisted modernization, he said. Investing in education and training for developers and engineers is also important, particularly as it relates to accelerating the AI learning curve.

“As we started the process, we weren’t nearly as effective as we were days, weeks into it, where we started breaking things into smaller components and learning how to ensure we had things properly documented, our prompts were correct, all of those things,” Bowen said. “So really learning that collaboration piece between the engineers and the AI tech [was important]. But that being said, it was pretty quick especially as it related to its ability to efficiently explain program usage, which is huge, especially when you’re modifying existing software. That’s really one of the key drivers from a time standpoint and quality standpoint.”

Bob also helps Medhost accelerate its design experimentation and prototyping stages. The company has always tried out different techniques, but with Bob, the development team can generate proof of concepts in minutes which previously would have taken hours, Bowen said.

“What we found with our usage of Bob as it’s been the most beneficial AI code assist tool for our RPG and IBM footprint than any other AI tech tool we’ve experimented with,” he said.

You can watch Bowen’s interview with Will here.

RELATED STORIES

IBM Gets Bob 1.0 Off The Ground

Bob More Than Just A Code Assistant, IBM i Chief Architect Will Says

IBM Pulls The Curtain Back A Smidge On Project Bob

Big Blue Converges IBM i RPG And System Z COBOL Code Assistants Into “Project Bob”