What the work actually is

You're producing and grading the raw material coding models learn from. In practice that means being dropped into an unfamiliar repository, tracing execution across services, libraries, APIs and databases until you understand how it actually fits together, then either constructing a task with a defensible ground-truth solution or judging whether a model's patch is correct. A large slice of the work is reviewing AI-generated diffs: does the patch fix the stated issue or only silence the symptom, does it respect the project's existing architecture, does it introduce a security or maintainability problem a reviewer would block in a real PR. Test authoring is part of the deliverable — unit, integration, regression and end-to-end — because a task without tests that distinguish a real fix from a plausible one has no evaluation signal. One named workstream involves translating complete builds from one language into another, so breadth across languages matters more than mastery of one.

What the screen is looking for

The bar is stated bluntly in the listing: contributions that show engineering, not documentation edits, dependency bumps, formatting passes or typo fixes. The AI interview runs roughly 30 minutes and is built around your own repositories — expect to be asked to name a specific PR, explain why the codebase was structured the way it was, what you changed, what broke, and how you knew it was fixed. Follow-ups go one or two levels deeper than the first answer, which is where people who describe work they only reviewed tend to come apart. Relevant languages include TypeScript, JavaScript, Python, Java, Go, Rust, C++, C#, Ruby and PHP; what's being tested is whether you can read an unfamiliar codebase and change it safely, in whatever language it happens to be written.

Logistics and pay

Fully remote and global, contractor status, roughly 15 hours a week with hours entirely of your choosing including weekends. Pay is observed at $100–150/hour, but note the structure: compensation is output-based and paid per accepted task, so the effective hourly rate depends on how quickly you work and how consistently your submissions pass spec review. Minimum weekly task submission requirements apply. Hiring moves fast — micro1 says roles typically fill within 48 hours, with first tasks expected 24–48 hours after onboarding.

  • Process: screening questions → ~30 minute AI interview → hiring manager review
  • Your GitHub profile must be public or otherwise verifiable
  • Be ready to discuss repositories, issues and PRs you personally authored