Most AI jobs are not engineering jobs
The fastest-growing AI roles involve judging language, not writing code. Models are trained and corrected by large numbers of people who read output and decide whether it is accurate, safe and useful. If you can read carefully and follow a detailed guideline document, you are qualified for the entry tier of this market.
The roles, translated
The titles vary between vendors, but the underlying work falls into five buckets.
- Data annotator — labelling text, images or audio against a rubric. $18–26/hr.
- AI trainer / RLHF rater — ranking model responses by quality. $20–35/hr.
- Evaluator / red teamer — deliberately probing a model for failures. $25–45/hr.
- Subject-matter expert — reviewing output in your professional field. Often $40/hr+.
- Prompt engineer — designing and testing production prompts. Usually salaried.
What the work is actually like
Expect a long guideline document, a qualification test, and then task batches with accuracy audits. Pay is often per task or per hour with a minimum accuracy threshold. The work is genuinely remote and genuinely flexible, but it is also detail-heavy and quiet — people who need variety and social contact often find it draining.
How to get hired
Vendors screen almost entirely on a qualification test rather than a resume. Read the guidelines twice before starting, take the test when you are fresh, and treat borderline cases by asking what the guideline's intent is rather than what feels right. Existing domain expertise — nursing, law, teaching, a second language — moves you into the higher-paid specialist tiers immediately.
Where it leads
Strong raters move into quality auditing, guideline authoring and project leadership, which are salaried roles with genuine career progression. It is one of the few remote functions where twelve months of contract work reliably converts into a permanent offer.
