How Unyte streamlined multi-repo development and unlocked team-wide AI expertise

Challenge

Unyte’s product spans five large, interconnected codebases, and key features often touch several at once. The prior AI coding tool struggled to navigate this scope, frequently reimplementing logic that already existed elsewhere rather than reusing.
Frequent releases across multiple environment-specific branches added coordination overhead to the deploy process.
Validating requirements and acceptance criteria for each feature required significant manual effort from product management and business analysis teams.
The existing tooling included access to Anthropic models, but usage limits couldn’t keep pace with daily development needs.

Drawing on experience with large-codebase migrations and AI-assisted modernization, Proxet recommended moving to Claude Code, with expanded access for developers most engaged in day-to-day coding.
Proxet led an all-team session–informed by Anthropic’s Academy–to walk through capabilities and work through questions and use cases in real time.
Produced a migration document covering account setup, developer environment configuration, and a forward-looking roadmap for shared skills libraries and MCP server integrations, for the team to keep expanding what Claude does for them after go-live.

Using the new tooling, the team planned and executed a payment system upgrade spanning two repositories and two programming languages — work that would have been far riskier to coordinate manually across codebases. The upgrade lays the groundwork for more flexible payment options and stronger customer retention.
A 31% increase in resolved aged backlog items–issues that had sat untouched for 3+ months–freeing engineering time previously spent context-switching onto legacy problems and redirecting it toward new feature delivery.
Bug fixes now pass QA review on the first attempt 37% more often, reducing rework cycles and shortening the time from bug report to resolution.
Built a custom skill that checks ticket requirements against epics, related documentation, and the codebase itself, catching gaps before development starts.
Developed a vision-enabled AI testing framework covering both functional and UX checks–surfacing visual and usability issues earlier in the release cycle, before they reach customers.