United StatesTechnology and telecom2024Amazon
Java upgrades with Amazon Q Developer code transformation
Amazon moved tens of thousands of internal Java applications to Java 17 with its AI coding agent
Problem
Amazon ran tens of thousands of production applications on Java 8 or 11. Upgrading one application by hand took developers weeks, and migrating a single dependency could take a day or more.
Approach
In 2024 Amazon used the code transformation agent in Amazon Q Developer, its own generative AI coding assistant, to upgrade internal applications to Java 17. Developers could review and adjust the agent's upgrade plan before it made the changes. Amazon sells the same tool to AWS customers.
Results
Takeaway
Amazon saved work on a large, repetitive maintenance task with a clear completion criterion. The savings are Amazon's estimates, with no independent measurement published.
Categories
- Size
- Large enterprise
- Industry
- Technology and telecom
- Business function
- IT and R&D
- Innovation type
- Process innovation
- AI technology
- Generative AIAI agents
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
AWS describes how Amazon used the Amazon Q Developer transformation agent to move tens of thousands of production applications from Java 8 or 11 to Java 17. It credits the agent with saving over 4,500 years of development work for more than a thousand developers and with $260 million in annual cost savings. It notes that a single dependency migration could take a developer a day or more by hand.
The Decoder reports Andy Jassy's statement that Amazon upgraded more than half of its Java production systems in six months, cutting the average upgrade from about 50 developer-days to a few hours. It gives his estimates of 4,500 developer-years and $260 million saved and says developers shipped 79% of the AI-generated code reviews without changes.
Cloud Wars quotes Andy Jassy saying Amazon migrated over 30,000 Java applications in a few months with Amazon Q's code transformation. He estimated savings of $260 million and 4,500 developer-years compared with doing the work otherwise.
Full details are in the original sources.