What the work actually involves

This is not a product engineering job. You are producing training and evaluation data for frontier models working on frontend problems. In practice that means authoring realistic tasks — a component with a subtle re-render bug, a layout that breaks at a specific breakpoint, a state management refactor with hidden coupling — then writing the reference solution and the criteria a grader would use to judge a model's attempt. Other project types lean toward review: you read model output, decide whether the JSX actually compiles and behaves as claimed, and write a rubric-anchored justification. Some engagements involve side-by-side preference ratings across model responses, where the hard part is articulating why one answer is better rather than just picking it.

Quality bars are stricter than most engineers expect. Reviewers check whether your task is actually solvable, whether your test cases discriminate between a correct and a plausible-but-wrong answer, and whether your written reasoning would hold up if someone disagreed with you. Vague feedback like "code could be cleaner" gets rejected.

What the platform screens for

Mercor runs an AI interview before you enter the network. It probes for verifiable specifics from your own work — which framework, which version, what broke, what you measured. Expect follow-up questions that go a layer deeper than your first answer, particularly around rendering behavior, state management tradeoffs, and performance profiling. Generic framework knowledge that reads like documentation tends to score poorly; concrete debugging stories with numbers tend to score well. The screen also assesses written communication, since nearly all deliverables are written English.

Logistics and pay

  • Fully remote, async, and self-scheduled; there are no standing meetings on most projects.
  • Typical commitments run 15–30 hours per week, with project length varying from a few weeks to several months.
  • Observed rates for this network fall in the $70–150/hr range depending on project, seniority, and client. Rates are set per engagement and are not guaranteed by joining the network.
  • Approval into the network does not produce immediate work. Matching is rolling and depends on what labs are requesting; some approved experts wait weeks before a relevant project opens.