Six AI Roles Hiring Now (and How to Get In)

Three years ago, if you told someone you were a “prompt engineer,” they’d ask if you worked for UPS. Then Anthropic posted a “prompt engineer and librarian” role at $175,000 to $335,000 and the job title went from punchline to career path in about a week.

The story everyone tells about AI is about displacement — which jobs will disappear. That’s worth taking seriously. But there’s a quieter, more useful story: the roles being created or reshaped right now for people willing to learn a new skillset. Indeed Hiring Lab confirms jobs mentioning AI kept growing through early 2026 even as overall hiring weakened.

One honesty note before the list: four of these six roles genuinely didn’t exist before the LLM wave. Two of them, data labeling and product management, are older jobs that AI reshaped into something new. Both kinds are hiring.

1. Prompt engineer

What you actually do: You’re not writing prompts all day like a copywriter. You’re architecting how systems interact with large language models: designing prompts that work at scale, testing edge cases, documenting what breaks, and often building evaluation frameworks to measure quality.

Pay: Glassdoor’s aggregate puts the US average around $130,000, with most reported salaries between roughly $102,000 and $166,000 (self-reported data, 2026). Specialized roles at top labs go higher — the Anthropic posting above topped out at $335,000.

Who’s hiring: The big labs (OpenAI, Anthropic, Google, Cohere) plus enterprises building domain-specific applications: banks, consultancies, hospital systems.

How to get in: You don’t need a CS degree. You need to demonstrate that you understand how language models behave and can write clear, testable specifications. Build a portfolio: take a workflow you know well, design and document prompts that improve it, show before/after results, and publish the write-up.

2. AI trainer (RLHF specialist)

What you actually do: You teach models to be better by grading their outputs. You evaluate responses, label data, identify when outputs miss the mark, and help models learn from human judgment. RLHF stands for Reinforcement Learning from Human Feedback; in practice, it’s quality assurance with consequences.

Pay: Varies more than any role on this list. Contract platforms pay hourly and rates climb steeply with domain expertise (code, law, medicine); full-time trainer roles at the major labs are salaried positions. Check live aggregates on Glassdoor or ZipRecruiter before negotiating — this market moves fast and published averages go stale in months.

Who’s hiring: OpenAI, Anthropic, Google DeepMind, Meta AI, Scale AI, plus companies building specialized AI for healthcare, finance, and legal services.

How to get in: Start with contract work at Outlier AI or similar platforms; the bar is accuracy and thoughtfulness, not credentials. If you have domain expertise — you’re a lawyer, coder, teacher, or scientist — that’s your edge. These companies pay more for trainers who actually understand the domain being graded.

3. Red teamer (AI safety specialist)

What you actually do: You try to break AI systems, systematically. You probe for biases, find failure modes, test adversarial inputs, document risks, and help teams build safer systems. It’s ethical hacking applied to AI.

Pay: Tracks security engineering, which is to say: well. Public postings at the major labs list six-figure ranges, with senior safety roles among the better-paid jobs in the field. Listings aggregated at aisafety.com/jobs usually include the posted range.

Who’s hiring: OpenAI, Anthropic, Google DeepMind, Redwood Research, plus government-adjacent contractors like Booz Allen and RAND.

How to get in: You don’t need a PhD, you need a critical mind and the ability to think like an attacker. Read papers on adversarial examples, practice on CTF (Capture the Flag) security challenges, document your attempts to make models misbehave, and connect with the AI safety community. Some organizations explicitly hire entry-level for this.

4. AI product manager

What you actually do: You build products powered by AI. You define what the AI should do, how users interact with it, when to fall back to traditional UX, and how to handle failures gracefully. Translating “we have an LLM” into “users get better search results” is a different skill than classic software PM work.

Pay: Tracks senior product management at tech companies, with an AI premium at the top end. Glassdoor’s AI product manager aggregate shows current self-reported ranges by level and location.

Who’s hiring: Every major tech company, every lab, and most Series B+ startups with an AI component.

How to get in: A PM background helps, but what matters more is showing you understand AI’s capabilities and limits. Write publicly about AI products. Audit existing ones — what works, what feels broken, why? If you’re at a company already, become the person who champions an AI project and ships it.

5. AI content safety evaluator

What you actually do: Quality assurance with ethical teeth. You evaluate AI-generated content for safety, bias, misinformation, and harmful reasoning. You’re not building AI; you’re grading its behavior and flagging what’s wrong.

Pay: The widest spread here — many positions are remote, part-time, or contract, and advertised rates range from standard content-moderation wages to specialist hourly rates for domain reviewers. Read the posting carefully; “evaluator” covers everything from gig work to staff roles.

Who’s hiring: Data-work platforms, content moderation contractors, and in-house trust-and-safety teams at the labs.

How to get in: The lowest barrier to entry on this list, and one of the most realistic starting points if you have zero technical background. No degree required; critical thinking and attention to detail matter more. Content moderation experience pivots naturally into AI safety evaluation. Apply directly through job boards.

6. Data labeler (specialized)

What you actually do: You create the labeled datasets that train models. The commodity version of this job is old; the specialized version is new. You annotate complex data, provide structured feedback, and often work in expert domains: medical imaging, legal documents, code.

Pay: Entry platform gigs pay modest hourly rates; specialized domains pay multiples of that, because a doctor labeling medical data isn’t replaceable by a generalist.

Who’s hiring: Scale AI and similar platforms, medical research institutes, legal tech companies, synthetic-data startups.

How to get in: The most accessible entry point in AI work. Start with contract gigs, build a track record of accuracy, and lean on any domain expertise you have. Most hiring here asks “can you do the work accurately?”, not “what credentials do you have?”

What these roles have in common

They’re not asking for PhDs, yet. Every one of these roles hires people without advanced degrees, asking for domain knowledge or demonstrated ability instead. That window probably won’t stay open forever.

Demonstration beats credentials. A portfolio of real work matters more than a certificate. An online course in prompt engineering is worth a fraction of a public repo showing your actual prompts and results. (The full argument is in Stop Collecting Certificates. Start Building Things.)

Adjacent skills are undervalued. Teachers understand learning, which is most of RLHF. Lawyers are gold for safety evaluation and legal data work. Coders are valuable everywhere. Flip your existing expertise into AI work instead of starting from zero.

These are builder roles. Nobody hands you projects. You experiment, document, and ship. If you have that instinct, you’re most of the way to hireable.

Remote-first is the default. Almost all of these roles are fully remote or offer it. If you’re outside the big tech hubs, you’re competing globally for jobs that pay accordingly.

Get started this week

You don’t need to go back to school, and you don’t need to quit your job to test the water. For prompt engineering, optimize the AI workflows at your current job and document the before/after — that’s a portfolio piece, and small freelance prompt gigs can supplement it (a side income while you build, not a career by itself). For training and labeling, apply to a platform like Outlier or Scale and put in real hours; accuracy is the reference that gets you the next tier. For safety work, read one paper on adversarial attacks, attempt a documented jailbreak, and write it up. For product management, audit three AI products and publish 500 words on what’s missing.

These roles are new enough that the “perfect candidate” doesn’t exist yet, and the people hiring know it. They’re looking for people who are curious, careful, and able to move fast.