KeyBank says Google's AI cut a 600-hour code migration to under 13 hours.
Google Cloud posted a film on 29 October about KeyBank, a US bank with about 950 branches and more than 200 years of history. Azim Shaik, who designs its technology as lead enterprise architect, says the bank carries "millions of lines of legacy code," old software still in daily use.
One engineering team used Gemini, Google's AI model, to build a tool that moves the bank's old programs onto newer systems. That job "usually takes 600+ hours," Shaik says. "We are now able to do the same thing in less than 13 hours." Google's case study adds that building the tool took five days instead of three months.
The bank also says teams now get new computer systems, already set up to meet the bank's rules, in minutes instead of weeks, a "more than 90%" cut.
The figures are KeyBank's and Google's own. Neither the film nor the case study gives a cost, a licence price or an outside check, and the result covers one team's job, not all of the bank's software.
Neither source says whether the 13 hours include testing that the moved code still gives the same results. In your legacy projects, is that testing still done by hand?
🎥 Video: Google Cloud, official YouTube channel, 29 Oct 2026 · narration: AI voice
What the captions say
- KeyBank runs millions of lines of old code.
- Moving one code base took more than 600 hours.
- A team used Google's Gemini AI to build a tool.
- The same move now takes under 13 hours.
- Setting up servers fell from weeks to minutes.
- Those are KeyBank's and Google's own figures.
- No cost and no outside audit were published.
