How Knowunity doubled engineering velocity with cloud agents
Knowunity is an AI learning platform used by more than 30 million students. Its AI tutor gives students step-by-step help with homework and exam preparation while turning their course materials into personalized study plans, flashcards, quizzes, and practice tests. Students can also learn from more than 1.6 million study notes shared by the community. The engineering team uses Replicas to run cloud coding agents across GitHub and GitLab and track how much work those agents ship.
Visit Knowunity“Replicas completely changed the way we run engineering at Knowunity by allowing engineers to trigger cloud agents from anywhere, like Slack, Linear, or with automations.”
2.8k
cloud agents launched in July
50%
of pull requests produced by cloud agents
2×
engineering velocity
2/3
less time spent fixing bugs
Turning background agents into measurable output
Knowunity wanted background agents to help the team ship faster and become a measurable part of engineering output. Engineering leadership set a concrete target for the first month: background agents should produce 25% of the team’s pull requests and merge requests.
Each agent also needed a full-powered development environment. Knowunity’s years of Git history, dependencies, developer tooling, and local services already fit inside a single workspace, and that setup needed to be ready for prompting in seconds. Replicas moved the team onto 4-vCPU, 16 GB RAM, 32 GB workspaces backed by prepared snapshots and a warm pool.
One workspace across GitHub and GitLab
Knowunity’s full-stack work crossed providers, with repositories in both GitHub and GitLab. Replicas put both repositories in the same workspace, so one agent could trace a change across the app and backend, edit both codebases, run the local stack, and open the corresponding pull requests and merge requests.
Engineers could trigger those workspaces from Linear, Slack, GitHub, GitLab, or the Replicas dashboard. Knowunity also encoded internal practices as agent skills, so agents could follow the team’s conventions wherever work started.
Cloud agents now produce half of all pull requests
In July, more than 30 Knowunity engineers launched 2.8k cloud agents through Replicas and logged more than 400,000 minutes running coding agents across thousands of GitHub pull requests and GitLab merge requests.
Cloud agents now produce 50% of Knowunity’s pull requests, doubling engineering velocity and reducing bug-fixing time by two thirds.
Replicas analytics gives Knowunity’s leadership a daily view of workspace usage, compute time, token spend, and PR and MR output by source and engineer. It also shows which agent skills are being used and which have the biggest effect on output, helping Knowunity compare token spend with engineering output and decide how to optimize agent use internally.
