As a Task Author, you will design and calibrate authentic engineering challenges — diagnosing failing integrations from logs, tracing bugs through codebases, and writing and verifying real fixes — that genuinely push the limits of capable AI agents.

This is hands-on engineering work. Generalist or theoretical profiles cannot do this. If you're a working software engineer who debugs real systems for a living, this is a rare opportunity to shape how the next generation of AI understands engineering.

Why Join Us

• Work on cutting-edge AI projects alongside leading research labs.
• Fully remote and flexible — work when and where it suits you.
• Freelance autonomy with the structure of meaningful, well-defined task-based work.
• Contribute directly to how AI systems understand and perform real software engineering.
• Potential for ongoing work and contract extension as new projects launch.

• Author realistic engineering task prompts covering integration failures, log-based debugging, bug traces through source code, and configuration or integration fixes.
• Write scoring rubrics that define exactly what a correct fix or diagnosis looks like and how it is verified — including live environment validation where applicable.
• Set up task environments with realistic codebases, logs, alerts, and system states that place an AI agent in a plausible engineering situation.
• Solve each task yourself — write the actual fix and confirm it works — to validate the task is sound and the rubric is accurate.
• Calibrate task difficulty by adjusting code complexity, log volume, failure modes, or ambiguity until the task reliably challenges the model to the intended degree.
• Review and correct AI-drafted task prompts or rubrics when provided.

• Working software engineer with hands-on, industry-level experience — this is not a theoretical or instructional role.

Work arrangement: Fully remote

Eligible countries: United States