How Parrot reviews every pull request with a team of cloud agents
Parrot is the TikTok for language learning. Its iOS app teaches Spanish through comprehensible input: short videos from native speakers, speaking practice, and vocabulary review sequenced to each learner. Parrot is used by more than 440,000 learners. Replicas is used by the team to review pull requests using teams of cloud coding agents and automate repetitive work across its TypeScript monorepo.
Visit Parrot“I think Replicas makes it super easy to spin up Claude coding agents and switch between different harnesses, which is very important to stay with the latest models. You get to use your same coding agents that you use normally, like Codex and Cursor and Claude, in the cloud, in a harness you’re very familiar with. Developer machines and cloud machines basically have the same environment now.”
23.5k
automated cloud-agent jobs in July
270k+
minutes running coding agents in July
Distilling review points into coding agents
Parrot ships quickly from one TypeScript monorepo. As coding agents became the team’s default way of shipping code, one of the biggest bottlenecks became code review. Compilation checks and linting didn’t cover everything, so the team also wanted every pull request checked for duplicated logic, unused code, unnecessary TypeScript casts, avoidable React effects, and API changes that could break existing users.
Replicas let Parrot turn each standard into a dedicated cloud agent. When a pull request opens, multiple automations run in parallel, each in its own virtual machine. One reviews the change as a whole. The others focus on a single class of risk, then post a comment or GitHub Check back to the pull request. Another agent matches PRs to Linear tickets and keeps their implementation status current.
Beyond code review
Parrot then extended the same pattern beyond code review. A migration agent starts the candidate backend against a fresh database fork, checks health, migration state, indexes, and lock contention, and tests whether a non-transactional migration survives a restart. A Slack automation turns bug reports, including screen recordings, into deduplicated Linear tickets with the evidence attached. Another agent sends the team a daily report on iOS release cadence and bug health.
Engineers also call Replicas directly from Slack and Linear. In July, they used it to implement product and growth features, investigate bugs, prepare App Store releases, and handle partner operations. The cloud platform became a central place for both autonomous checks and interactive work.
The outcome
In July alone, Parrot ran 23.5k automated cloud-agent jobs through Replicas and logged more than 270,000 minutes running coding agents.
Parrot now has a simple way to spin up background agents for all types of software engineering work, like code review, bug fixes, and feature building. With Replicas' shared environments and Slack integration, it served as a very easy way for non-technical people on the team to code up prototypes or ideas and share with with other people.