AI & the future of work — answered straight.

In 2026, roughly 56% of layoffs cite AI as a factor and over 150,000 workers have been affected — but the picture is messier than the headline. Klarna, for example, publicly rehired people it had cut. AI is removing the value of narrowly-scoped, repetitive tasks fastest, not entire professions overnight. The real risk is being the person who only does the task AI now does for free; the real opportunity is becoming the person who directs, judges, or fixes what AI produces in your field. See Issue 6: The Layoff Map for the full breakdown.

It's the human work behind the AI you use every day: writing example prompts, rating model responses, catching mistakes, and testing edge cases — done by people with real domain expertise (lawyers, nurses, teachers, engineers, writers), usually through platforms like Mercor, Outlier, and Handshake AI rather than by an AI lab's own employees. See the AI Platforms Directory for a vetted list.

In the United States, almost all AI training, evaluation, and AI-tutor work — through platforms like Mercor, Outlier, Handshake AI, Alignerr, and Scale AI — is structured as 1099 independent contractor work, not W2 employment. Practically, that means: no employer-provided health insurance, no employer-paid share of payroll taxes, no PTO, and you're responsible for self-employment tax (about 15.3% on top of regular income tax) plus quarterly estimated payments to the IRS.

In exchange, 1099 status usually means more flexibility — no fixed hours, no manager, the ability to work across multiple US platforms at once — and the ability to deduct legitimate business expenses (a portion of your internet, a home office, equipment) that a W2 employee can't.

A small number of roles, mostly full-time positions at the AI labs or platforms themselves rather than gig-style evaluation work, are W2. If a listing doesn't say which one it is, assume 1099 until a formal offer says otherwise — and don't treat a platform's advertised hourly rate as your take-home pay without running your own numbers on self-employment tax.

This is general information for US workers, not tax advice. Talk to a US tax professional about your specific situation, especially once your AI income adds up across multiple platforms.

No. Most demand right now is for domain expertise, not code — a nurse rating medical-reasoning answers, a lawyer testing contract-review accuracy, a teacher evaluating tutoring responses. Coding-focused evaluation work exists too, but it's one lane among many, not a requirement to get started. See Issue 2 for a real example of domain expertise turning into paid AI evaluation work.

Published rates across platforms range roughly $15–$150/hr depending on the platform and how specialized your expertise is; real earners have reported blended rates in the $80–90/hr range across a mix of projects — see Issue 8 for a real earnings screenshot and contract breakdown. It's part-time, project-based income rather than a guaranteed salary replacement, and availability varies week to week.

Yes — the reputable ones are backed by real venture funding and have major AI labs as paying clients. That said, they're gig platforms: no guaranteed hours, project-based availability that ebbs and flows, and the normal due-diligence rules for any gig-work platform still apply.

Training means creating or labeling the data used to teach a model in the first place. Evaluation means judging a model's output against a rubric after it's built. Red-teaming means deliberately trying to break or misuse a model to find its weaknesses before a bad actor does — see Issue 7 for the red-teaming opportunity in more detail.

Depth in your existing field, applied to AI, beats a generic prompt-engineering certificate almost every time — see Issue 5 for the role-by-role breakdown. The highest-paid work goes to people who can tell when an AI answer is subtly wrong in their domain, not people who can write a clever prompt.

An LLM is the underlying model that predicts text; an Agent is that model given a goal and the ability to act on it in steps; MCP is the standard that lets an agent connect to outside tools and data. See the full FOWL AI Glossary for LLM, Agent, MCP, Skills, and Claude CLI — each defined in one sentence with a real-world example.