SoundHound AI remediated 6,000 issues across 45 codebases with automated recipes.
SoundHound AI has remediated more than 6,000 issues across 45 codebases with Moderne, completing Java 21 and Spring Boot migrations estimated to save more than 1,700 hours. Engineers also run Moderne Prethink so their coding agents start with the codebase already mapped.
- Technology
- Industry
- 25M+
- Lines of code
- 300+
- Repos
Security fixes at SoundHound AI no longer compete with feature work for attention.
The team estimates the automation saves 30 to 50 percent of the time its remediation tasks used to take. Those are now hours developers can spend building features.
We’ve been able to achieve stylistic standardizations across the organization, which has been helpful.
SoundHound AI turned its fixes into recipes that run in its pipelines.
The team started with vulnerable dependencies and formatting, then moved to larger migrations such as from Spring Boot 2 to 3 and from earlier Java versions to Java 21. SoundHound AI’s engineering teams work across time zones, so a fix that touched many repositories used to wait for each owning developer to come online and make the change by hand. Fixes are now written once as recipes and run in the company’s CI/CD pipelines, which apply them to every affected repository without waiting on anyone.
SoundHound AI has written more than 30 recipes of its own, which run alongside the public catalog and cover changes specific to its code. One of them started the Spring Boot 4 migration in a single run across the whole codebase. Their previous Spring Boot migration took more than six months and several engineers.
We could address a bunch of our technical debt along with the vulnerability reporting. It quickly became evident that Moderne was much better at doing a lot more than just basic vulnerability scanning.
SoundHound AI now fixes the vulnerabilities its scanner used to leave for developers.
SoundHound AI builds conversational AI for enterprise contact centers and handles billions of customer interactions a year. Its engineers maintain more than 300 repositories and 25 million lines of Java, Spring Boot, Python and JavaScript.
Veracode flagged vulnerabilities but left the fixing to developers, one repository at a time. The work was non-billable, so every hour came out of engineering output, and technical debt kept building. Ahead of its Veracode renewal, the team went looking for a way to fix what the scanner found.
With Moderne, SoundHound AI can fix what the scanner flags across every affected repository, and engineers review the changes.
The Moderne team has been extremely responsive and engaging, definitely more than Veracode by an order of magnitude.
Prethink gives SoundHound AI’s coding agents context before they start.
SoundHound AI’s engineers use coding agents, which need to know what the code contains before it makes any changes. On a large estate, that discovery is most of the work. Moderne Prethink does it ahead of time deterministically, using zero tokens. A recipe gathers data from the LST, including dependencies, architecture, test gaps and more, and writes it into the repository so the context is already there when the agent opens it.
The team is bringing Prethink further into its pipelines. In one month, two Prethink runs covered 197 repositories.
There really isn’t another product doing what Moderne is doing. The ability to find things across a large set of codebases is invaluable.
Related resources
See deterministic, estate-wide code change on your own repositories.