We're looking for experienced software engineers to build AI coding tasks from a codebase they know well. US$1,000 per task, and you can do as many tasks as you want. You turn a real change from a repository you know into a task that today's best AI coding models struggle with: hard for the models, but fair for a skilled engineer. Your tasks are used to measure and train AI coding models. Remote, any location. US$1,000 per task that passes review. Minimum of one task a week, about 10 hours a week. How it works Apply. Complete a short screener interview and upload your resume. Propose. Once you're selected, pick a change from a repository you know and submit a short proposal. We approve it or suggest a better candidate. Build. In our platform, set up the repository as it was before the change, then write the brief, the correct solution and the grading checks. Stress-test. Try lazy, incorrect and shortcut solutions and make sure your checks catch them. Model testing. Run your task against leading AI models until strong models often fail but a skilled engineer can clearly solve it. Review. A reviewer checks your task and you revise it. Most tasks take more than one round. Next task. Once your task is accepted, start another. Requirements 2+ years of professional software engineering Code merged into a public open-source repository, ideally one you know well. A private codebase you own may also qualify (not employer or client code) Strong in at least one of Python, TypeScript, Go, Rust, Java or C++ Solid with Docker, Linux, Git and automated testing Regular use of an AI coding tool Experience writing tests, rubrics or evaluation criteria, ideally for AI models You can explain a change clearly and concisely, in your own words At least 10 hours a week, starting now A secure personal or work computer (no public or shared machines) Nice to have Deep expertise in one area, such as databases, compilers, distributed systems, security or infrastructure Experience building agent