Profound Says New Agentic AI Dev Tool Delivers Huge Productivity Boost
January 19, 2026 Alex Woodie
Profound Logic is rolling out a new AI-powered development tool that it says will increase the productivity of individual developers by five to 10 times. Dubbed CoderFlow, the software coordinates the work of multiple AI agents to assist with the development new IBM i applications or the maintenance and modernization of existing ones.
Existing AI copilots provide some productivity gains for developers working in an IDE, somewhere around 20 percent, according to Profound Logic chief executive officer Alex Roytman. But CoderFlow is a different type of product. It is not a coding co-pilot, he said. Instead it is a framework for spinning up autonomous agents that can accomplish a range of tasks, from converting code to new languages and compiling the code to testing code in a 5250 or Web environment and even resolving errors.
“What we’re seeing with this new approach is it’s truly agentic and it’s like five to ten times the productivity,” Roytman told IT Jungle. “I always feel like if I say something like this, the person on the other side is like, that sounds like BS.”
But CoderFlow is the real deal, Roytman insists.
Profound CoderFlow provides a range of agentic capabilities for IBM i developers.
One of the product’s advantages over traditional GenAI copilots is its multi-agent framework. Once the customer submits a given task to CoderFlow, such as converting an RPG subprocedure from RPG III to free-form RPG IV, the software will spin up multiple agents automatically to tackle the task. The software will run a competition among the agents – each of which may be using a different language model, such as Anthropic’s Claude or OpenAI’s GPT 5 – to see which one will produce the best results.
“There’s another agent that will look at all of those and say, out of all of these, here’s how I rate them,” Roytman said. “It’ll produce a little rubric and say: ‘This one got the functionality right, but the code is not as well organized as it could be, whereas this one’s got the code well organized.’ Because they’re competing, the winning solution will come to the front.”
The agents have their own containerized environments to work in, which gives them autonomy and also limits the damage they can do if they do something wrong, Roytman said. If an agent brings up a 5250 session to test a change, it’s a virtual emulator, not a real one.
CoderFlow understands and can generate a variety of languages common the IBM i platform, including various flavors of RPG, COBOL, CL, DDS, SQL, Java, and Node.JS; it hasn’t been tested in C on IBM i. Whereas the mainstream LLMs like Claude and GPT struggled with RPG just a year ago, they have made rapid improvements and now display a solid of RPG, Roytman said.
Profound CoderFlow agents check their work after completing tasks.
Those improvements mean that any IBM i shop can go out and use Claude or GPT 5.0 to understand and generate RPG on their own. What CoderFlow brings to the table is something different, Roytman said.
“These models do understand the RPG fairly well,” he said. “We provide the scaffolding, the parts that will say, well, in order for me to understand more about this RPG program, in addition to just reading the code, I have to be able to understand the context. It’s going to go out and understand the data, the database tables that I’m working with, what the layout is. The agent can’t really make decisions on how to edit the RPG without knowing the right context.”
Like many GenAI sessions, a CoderFlow session starts with a prompt. A human developer may tell the software to build something, such as an order processing module, along with validation, error handling and API support. CoderFlow will then analyze the command, break it down into individual tasks, and then spin up AI agents to execute those tasks.
Interaction between the human developer and the software is critical for getting it right. The first set of results may be throw-away work and in need of improvement, Roytman said. The software allows customers to deliver feedback in a couple of different ways, including by circling elements on the screen or by explaining the errors in typed input, he said.
A customer may have dozens of Profound CoderFlow agents simultaneously working on tasks.
“It is kind of like a chat, where you say ‘You got this almost right, but can you adjust this part or can you adjust this piece,’ and it will make the adjustments for you,” Roytman said. “Once it’s edited the code, it then moves on to saying, well, let’s see if it compiles. It resolves all the compilation issues; let’s see if it runs. If it’s a 5250 application, it spins up a 5250 session on its on its own, runs through the application, test everything that it’s built. If there are runtime errors, it resolves them. If something doesn’t appear right on the screen, it resolves them. And it can do that with both browsers and 5250 screens.”
Instead of staring at an IDE all day, the developer is now coordinating the work of multiple agents. Having multiple agents compete to create the best solution bumps up the success rate, Roytman said. All told, CoderFlow can significantly improve the amount of work, whether its new development, maintenance of existing code, or modernizing older software.
You don’t have to be a coding genius to use CoderFlow, but it does require someone with moderate RPG skills, as well as an understanding of the database layout. There is a bit of setup required to use CoderFlow, with the biggest requirement being that a customer must be storing their source code in Git, and managing changes in Git as well, Roytman said. Setup and implementation will typically take a couple of weeks, potentially longer if the customer is not yet using Git.
CoderFlow is a component in Profound App Dev, the company’s flagship application development suite, which encompasses a variety of other products, including Profound UI, Profound.js, Genie, and Atrium. In addition to the Web interface, Profound is offering command line and VS Code extensions.
Roytman is confident that progress in agentic AI is going to provide big productivity gains for IBM i shops in 2026.
“That doesn’t mean, oh, let’s just start getting rid of people. I think companies are going to become more ambitious and start doing projects that they thought were out of reach,” he said. “You can do it much faster than you did before. You’re going to be taking on more work.”
RELATED STORIES
Profound Logic Adds MCP To IBM i AI Tool
