What the work involves

You'll produce and critique deliverables that look like the ones you'd send to a client or a deal team, not abstract Q&A. A typical task is a prompt plus a source agreement: redline the vendor MSA from the buyer's side and justify each change; build a diligence issue list from a data-room folder; turn twelve contracts into an obligations-and-renewals tracker in a spreadsheet; draft a five-slide negotiation position summary for a GC. Other tasks hand you a model's attempt and ask you to score it — did it catch the uncapped indemnity, did it flag the anti-assignment clause in a change-of-control context, did it invent a defined term that doesn't exist in the document.

The judgment that matters is transactional, not academic. Reviewers are expected to distinguish a genuine risk from a stylistic preference, to explain why a fallback position is acceptable at a given leverage level, and to notice when a model produces something fluent and confidently wrong — a plausible-sounding limitation of liability carve-out that contradicts the survival clause three pages later. Written rationale is part of the deliverable; a correct redline with a thin explanation is a weak submission here.

What the screen looks for

  • Verifiable ownership. Four-plus years reviewing and negotiating commercial agreements, with clauses you personally redlined and positions you personally held — not deals you staffed.
  • Depth under follow-up. The AI voice screen asks a clause question, then asks why, then asks what you'd accept instead. Answers that stop at the first level don't clear.
  • Artifact craft. Spreadsheet and slide fluency is a stated qualification, not a nice-to-have. Obligation trackers and issue lists are graded on structure, not just substance.
  • Evaluation instinct. Willingness to mark a polished-looking output as wrong, and to say precisely which sentence causes the problem.

Logistics

Fully remote and asynchronous, with work claimed from a queue rather than assigned on a schedule. Flexible 5–20 hours per week, more if you want it; contributors commonly work in evening or weekend blocks. Screening is an AI-led voice interview, so expect to speak your reasoning aloud rather than write it. Pay is observed at $100/hour and is not guaranteed across task types or over time.