What the work actually is

Despite the job title reading like a product role, the day-to-day is AI training work. You'll be handed frontend prompts — build a responsive data table with sortable columns, fix a focus-trap bug in a modal, restructure a component so state lives in the right place — and either write the reference solution yourself or judge two model attempts against each other. Tasks are usually self-contained and take anywhere from twenty minutes to a couple of hours. The deliverable is rarely just the code: it's the code plus a written rationale that a reviewer who didn't write it can follow.

A good portion of the queue is accessibility and semantics. Models are fluent at producing markup that looks correct in a screenshot and falls apart under a screen reader — divs acting as buttons, ARIA attributes applied to the wrong node, colour contrast that fails at the token level. Catching that reliably, and articulating the specific failure rather than 'this isn't accessible', is the skill that separates contributors who stay on the roster from those who don't.

What the platform screens for

micro1 runs an AI-conducted interview before any human review. Expect it to ask about frameworks you've named on your profile and then follow up one or two levels deeper — if you say React, be ready to talk about why a particular re-render happened, not just that memoization exists. It probes for concrete project detail: what you built, what the constraint was, what you'd change. Vague seniority claims without a specific artifact behind them tend not to survive the follow-up. Written English matters here more than in most engineering roles, because your written justifications are part of the training signal.

Logistics

  • Fully remote, contractor status, no fixed hours; work is claimed from a queue.
  • Asynchronous — occasional calibration sessions, but no standing meetings.
  • Volume fluctuates. Most contributors treat this as part-time supplementary work rather than a full load.
  • Rate placement within the $30–80 band reflects task complexity and review track record; the band is what's been observed, not a guarantee.