What the work actually involves

You take a realistic business task — reconciling a vendor invoice discrepancy, drafting a board update from scattered source documents, triaging a support backlog — and run it end to end through an AI assistant with tool access. That means connecting the model to Google Drive, Gmail, Slack or Notion where the scenario calls for it, then documenting every turn: what you asked, what the agent did, where it silently skipped a step, invented a file, or produced something that looks correct but no professional would send. You then score the transcript against a supplied rubric and write feedback that a research team can act on without having watched you work.

The volume of writing surprises people. A single scenario often yields more annotation prose than the task itself took to run, and reproducibility matters — someone else should be able to follow your record and hit the same failure. Recurring-pattern write-ups are valued highly: noticing that the agent consistently loses context after a tool call is worth more than flagging the same hallucination twenty times.

What the platform screens for

micro1 runs an AI-led interview. Expect it to probe how you personally use ChatGPT or Claude in your job — specific workflows, specific limitations you've hit, what you stopped delegating to the model and why. Generic enthusiasm reads badly; the listing is explicit that daily working use, not experimentation, is the bar. Rubric experience is the second axis: grading, QA scorecards, hiring rubrics, content review guidelines, annotation work. The third is written English under time pressure, which the screen tends to measure through the length and structure of your spoken answers as much as any writing sample.

Logistics

  • Contractor engagement, remote, within the US, Canada, UK, Ireland, Australia or New Zealand — US strongly preferred.
  • Largely asynchronous; you pick up scenarios from a queue rather than sitting in meetings.
  • Hours are flexible but batched — most contributors report committing to a defined weekly block rather than dipping in.
  • Rates observed in the $30–90/hr range, varying with domain, task type and project phase. Nothing here is guaranteed.
  • Bring your own accounts and workspace tools; you'll be connecting AI assistants to real productivity platforms.