IBM: Bob becomes a multi-agent platform — nine-month COBOL modernization project completed in 3 days

On July 9, 2026, IBM announced a major upgrade to Bob, its agentic environment for software development. Agentic development means that AI agents — not just autocomplete — plan, write, test, and review code; the new version of Bob introduces a multi-agent architecture in which specialized agents coordinately cover the full development lifecycle.

From a single agent to an orchestra

Key additions: multi-agent coordination across the development lifecycle, parallel tool calling natively in the model, and sub-agents in isolated environments that reduce token consumption. The new Bobalytics component gives management insight into productivity, code quality, and agent costs — responding to a finding IBM cites: 85% of DevSecOps professionals say the bottleneck has shifted from writing code to code review and validation.

Why the mainframe package is the most interesting part

IBM Bob Premium targets the segment where IBM has no competition: modernization of COBOL and PL/I code and JCL analysis on IBM Z mainframes, integration with IBM i systems, and migration to Java 25 with refactoring. Industry estimates have for decades spoken of hundreds of billions of lines of COBOL in production at banks and insurers — code that nobody fully understands and whose authors are retiring. Reference example: client Blue Pearl completed a nine-month modernization project with Bob in three days.

Market context

While GitHub, OpenAI, and Anthropic build agents for greenfield development, IBM attacks the opposite end of the spectrum — legacy systems where the value of modernization is greatest and tolerance for error is lowest. The announcement comes the same day IBM’s competitive set (GitHub Copilot with GPT-5.6, OpenAI ChatGPT Work) also raised agentic stakes, but with a very different audience: Bob is sold to executives who count mainframe MIPS, not developers on Twitter.