Working with Google Cloud engineers, the team built a structured agentic workflow, first mapping the legacy system, then testing new architecture options before committing to the rebuild. In just eight weeks, Nextory migrated 100,000 lines of Ruby code, a project estimated at around six or seven months.

The second phase was the four-day co-engineering week, an intensive in-person engineering sprint in which Google Cloud and Nextory engineers worked side by side. The teams used agents to build and compare Java, Python, and Go prototypes of the same business logic, and compared deployment options, including Cloud Run and Google Kubernetes Engine (GKE).

By the third day, the team had settled on a new architecture: Python for faster audio processing, Java for the ingestion model, Pub/Sub to decouple each stage of the workflow, Workflows to coordinate the process, and Cloud Run to scale processing capacity automatically. The result was not just a faster implementation, but an event-driven architecture that removed hard-coded dependencies, improved scalability, and gave Nextory a clearer model for future extendability.

“This project visualized the way we should work going forward. The migration itself was valuable, but learning to work with agentic AI and seeing how much more efficient we can be was huge,” says Engblom. “The Google Cloud team never just said, ‘Do this.’ They guided us through different options, helped us understand the skills, and showed us how to use them efficiently in our day-to-day work.” The company is now exploring the ways it can operationalize agentic development: what context agents need, at what stage, and for which product teams.

“We talk a lot about how much faster the process was compared to a traditional migration,” Engblom concludes. “But the important thing is that we gained that speed while maintaining quality—even improving it. It’s easy to do things fast, but we also built a system that works really well.”