The State of AI-Native Jobs

Two years ago, "AI role" mostly meant a researcher at a handful of labs. That era is over. Across the 48 AI-native companies we track, hiring now spans engineering, product, data, and research — and the roles share a common thread: they exist to turn model capability into something a business can depend on. AI has moved from a feature a few specialists worked on to the infrastructure entire companies are built around.

That shift shows up in the numbers. There are roughly 2,070 open roles in our index at any given time, and the bulk of them are not exotic research positions — they are engineering and product roles that ask for production experience with models, not a PhD. The market has broadened far faster than the supply of people who have actually shipped this work.

The result is a candidate's market for anyone who can demonstrate real capability, and a confusing one for everyone else. Titles are inconsistent, compensation is opaque, and the interview itself has changed shape. This report is an attempt to make the market legible: what it pays, where the demand sits, who is hiring, and what they actually test for — all computed from live postings rather than self-reported surveys.

A note on shape before the numbers: the market is weighted toward mid and senior individual contributors. That matters because it means the centre of gravity is reachable — most AI-native demand is at levels you can hit with a strong portfolio and solid fundamentals, not only at the frontier.

Compensation is the clearest signal of how much this work is valued, and the headline is a premium. The median AI-native role in our index carries a total compensation of $208k — roughly 28% above what a comparable engineer would earn on a traditional software team. The premium is not a quirk of a few outlier offers; it holds across the distribution, because at an AI-native company the work compounds directly into the product's core value.

The spread is wide, and it widens at the top. Research and safety roles lead — a small, scarce talent pool concentrated at well-funded labs — while the broad AI Engineer band anchors the middle and carries the most volume. The chart below ranks the highest-paying roles by median posted compensation.

The mechanism underneath the band is equity. At senior levels, the equity grant rivals or exceeds base, so two offers at the same title can differ by tens of thousands of dollars depending on stage and how the grant is structured. The headline total-comp figure is really a year-one estimate that depends on how the company performs over your vest.

For candidates, the practical implication is to read offers in components rather than as a single number. Base is what you can count on; equity is the upside you are betting on. The roles at the top of the chart pay the most precisely because they carry the most of that upside-weighted, scarce-skill compensation.

Pay tells you what a role is worth; open-role counts tell you where you can actually get hired. And demand is lopsided. A handful of generalist engineering roles — the ones that turn model capability into shipped product — account for a disproportionate share of all postings, alongside the data roles that feed those systems.

That concentration is good news for anyone choosing where to aim. The highest-volume roles are also the most flexible entry points: they hire continuously, span every company stage, and reward demonstrated capability over credentials. Niche roles pay more per head but hire in far smaller numbers, which makes them a slower path in.

The roles below are ranked by open postings in our index. The takeaway is not "chase the biggest number" — it's to pair a high-demand role with a scarcer, better-paid specialisation within it. Breadth of demand gets you in the door; depth gets you the offer.

Level is the single biggest lever on an AI-native offer — bigger than company, bigger than location. The same title spans a wide band, and the jump from mid to senior is where most of the compensation growth happens, because that is where the job changes from owning well-scoped tasks to owning systems end to end.

Above senior, bands widen further and become individualised. Staff and principal compensation is increasingly equity-dominated, so public ranges get noisier the higher you look — the number is more a bet on trajectory than a quote.

The lesson for candidates is to negotiate level before base. One level up moves base, equity, and refreshers together and compounds for years; a base negotiation at a fixed level moves a fraction of that. The table shows how comp scales across levels at the market median.

The busiest hirers are not a random list — they are the companies scaling a function, which has real consequences for candidates. High volume usually means clearer levelling, more structured interviews, more room to negotiate, and faster internal mobility once you're in. It is, bluntly, where the highest-probability offers are.

The most active AI-native employers in our index lead the table below. They aren't always the highest payers, though at this tier the two correlate closely; the frontier labs combine both large headcount plans and top-of-band, equity-heavy packages.

If you're early in a search and unsure where to point a focused, well-prepared application, the companies below are a pragmatic starting list — simply because they're hiring the most.

We parse the skills named in every live posting and rank them by frequency. Python is effectively universal — it shows up in the large majority of AI-native postings — which makes it table stakes, not a differentiator. If it's missing from your resume you're filtered out, but having it doesn't set you apart.

The real signal is in the long tail, and the fastest-rising skill there is evals: the ability to measure whether an AI system actually works. As teams move from impressive demos to production they need people who can answer "how do you know it's good?", and that question increasingly appears, explicitly or implicitly, across postings and interviews alike.

The strategy that follows is simple to state and hard to execute: secure the table-stakes skills, then go deep on one fast-rising differentiator. Breadth gets you past the automated filter; depth on a scarce skill gets you the offer and the salary lift that comes with it.

Remote roles have compressed — but not closed — the geographic gap. Roughly 40% of current AI-native postings are remote-friendly, concentrated in engineering and research; customer-facing and hardware-adjacent roles skew on-site or hybrid.

The Bay Area still leads on pay, with New York closing fast. But a strong remote offer now lands within striking distance once you adjust for cost of living — and for many candidates, optimising for realised, spendable income points to a remote role in a lower-cost metro over the highest nominal number in an expensive hub.

Put the pieces together and a clear picture emerges: AI-native hiring is broadening, paying a durable premium, and rewarding demonstrated capability over pedigree. The roles with the most volume are also the most accessible, the highest pay sits with scarce skills near the frontier, and the interview increasingly tests whether you can ship and measure real systems.

For candidates, the playbook follows directly. Target a high-demand role, go deep on one scarce, well-paid skill within it, build a portfolio that proves it, and prepare specifically for the AI-native interview loop. "Apply to more jobs" is the wrong optimisation; "be more clearly ready" is the right one.

We'll keep this report updated as the market moves. The numbers will shift — bands re-benchmark every few months — but the structural story is unlikely to reverse soon: AI is becoming infrastructure, and the people who can build and run it are still scarce.

Every figure in this report is computed from live job postings across AI-native companies, refreshed monthly. Where a company posts a band we use the midpoint; where it posts equity we annualise it over a standard four-year vest. Roles are normalised to canonical titles, and combinations without enough postings to be statistically meaningful are excluded rather than reported thin.

These are calibrated ranges, not offers — actual numbers vary with location, stage, performance, and the share price at each vest. For the full detail on how we compute everything, see our methodology.