What the work actually is

You are not doing admin work here; you are writing the exam that tests whether an AI agent can do admin work. A typical task starts from a scenario you invent from experience — a director's three-city trip collapses when a connecting flight is cancelled, and the agent must rebook, reconcile two conflicting calendars, notify four attendees with different seniority, and flag which receipts now fall outside policy. You then build the supporting artifacts: the email thread, the calendar exports, the PDF receipts, the contact records, the expense policy the agent is supposed to have read. Then you write the gold-standard solution as a sequence of decisions, and a rubric of 35 or more criteria that separates "technically completed the booking" from "handled it the way a seasoned EA would."

The rubric is where most of the difficulty lives. Process criteria are easy to enumerate — did it check the policy, did it cc the traveller. Judgment criteria are harder: did it escalate rather than guess, did it choose the right tone for the external vendor versus the internal team, did it notice the double-booking nobody flagged. Reviewers push back on rubrics that are all checkbox and no judgment, and on scenarios that are solvable in two steps.

What the screen looks for

  • Real tenure in a real org. Five-plus years supporting teams or departments in mid-to-large organizations is the stated preference. Expect follow-ups asking which systems, whose calendars, what volume.
  • Process ownership, not ad-hoc tasking. The listing specifically distinguishes recurring workflows you owned from assignments you were handed. Have two or three named processes you inherited, documented, and improved.
  • Multi-tool fluency. Scenarios span calendar, mail, expense, document, and contact systems, and artifacts come in CSV, PDF, and spreadsheet form. You will be asked how you'd represent a messy real-world dataset.
  • Writing. Professional correspondence standards are part of the product — the emails you fabricate have to read like real emails from real people with real politics.

Logistics and pay

Fully remote, asynchronous, contractor engagement. Compensation is output-based: paid per task that meets specification, with an observed effective band of roughly $20–40/hr depending on how quickly you work and how often your submissions pass review on the first pass. That band is observed, not guaranteed, and rejected or heavily revised tasks do not pay. A weekly minimum submission count applies. micro1 typically fills these roles within 48 hours and expects first tasks within 24–48 hours of onboarding, so treat the screen as an immediate-start conversation.