IBM Tested Bob on Itself First

IBM didn't launch Bob, its new AI coding agent, by handing it to customers first. It started with giving it to itself. Today, more than 100,000 IBM developers run Bob internally on everything, from decades-old mainframe code to modern cloud systems.

So what makes Bob different? Every chief data officer has a codebase still running payroll or the ledger, the one nobody wants to touch. Bob is built to work inside it, reading existing code and letting a developer check the logic before the AI touches anything.

IBM tested the rebuilt V2 the same way, running it across that internal workforce for weeks before the June 24 release. The upgrades address one worry: what happens once an agent works without someone watching every move. Bob can now run tasks in the background while a developer works on something else, and undo any AI-made edit down to the single action that caused it. IBM has also packaged the approach into playbooks for the projects that actually worry CDOs, like migrating old Java systems or modernizing platforms banks and insurers can't afford to touch. IBM's release notes for V2 spell out the mechanics.

Speed is the easy claim; every vendor in this category makes it. IBM says it designed the agent from the ground up designed the agent groundup for production access, with approval and rollback built in rather than added later. This means it is less like a coding assistant and more like a compliance product that writes code on the side.

Still curious about IBM Bob? Start your test drive here: https://bob.ibm.com/blog/

Image credit: iStockphoto/ismagilov