What AI companies have been hiring for
Most AI companies post their openings through Greenhouse, Lever, or Ashby, whose APIs return every role as structured data. That comes to 9,714 open roles across 102 companies. Their descriptions spell out the languages each role asks for and the kind of engineering behind it: training models, running infrastructure, or deploying to customers.
Numbers are a snapshot from 2026-09-10. “Live” means open on the board that day. Each ATS API returns only currently-posted roles, not a trailing window, so a job filled or pulled before the 10th isn’t counted. The set skews fresh. The median opening was first posted 60 days earlier, and 80% went up within the last six months. 102 of 156 companies expose a public board; the rest run Workday or a custom site this pass didn’t fetch, so big labs and defense primes are under-counted here.
What kind of engineer they need
Job titles hide the real signal. Of the 9,714 openings, 5,150 are technical, and each description is tagged for the kind of experience it asks for.
Building and training models is only 10% of technical roles. Most of the work sits around the models, not inside them. DevOps and cloud infrastructure lead at 38%, then customer-facing deployment and agent engineering. If an “AI job” brings to mind training neural nets, the market wants something else. It hires people who can deploy, scale, and wire models into products.
Experience mix by sector
Darker means a larger share of that sector’s technical roles ask for it. The split is sharp. Applied-enterprise companies (Harvey, Sierra, Decagon) are almost entirely about agents and getting deployed inside a customer, at 78% and 67%. Infrastructure and coding-tool companies want DevOps and distributed systems. Only frontier labs and robotics ask much for research or model-building, and even there it trails infrastructure work. Pick the sector that matches the engineer you want to be, not the one with the best logo.
Inside the technical roles
Those same 5,150 technical roles, split by specialty, say more than the raw headcount does.
Generalist “software engineer” and infrastructure/platform together are more than four in ten technical roles, while dedicated frontend, full-stack, and mobile barely appear. These companies hire broad systems engineers, not web specialists. And forward-deployed engineering is the third-largest specialty, ahead of applied ML, which is a recent shift in what “AI job” means.
Specialty mix by sector
The specialty split shifts hard by sector. Each cell is the share of that sector’s technical roles in that specialty (darker is higher).
Generalist software engineering is the base layer in every sector, heaviest in applied enterprise, silicon, and coding tools. The specialties around it are what move. Forward-deployed and solutions roles cluster in generative media, infrastructure, and coding tools, and nearly vanish in robotics and silicon. Research and applied-ML titles concentrate in frontier labs, generative media, and robotics. Robotics carries almost all the embedded work, while silicon leans on hardware-adjacent engineering that sits mostly outside this grid. Platform and infrastructure hiring is the near-constant, around a fifth of technical roles everywhere except generative media.
What they actually ask for
Stack mentions counted across the same 5,150 technical descriptions.
Python is in nearly half of them, no contest. After that the demand is infrastructure: Kubernetes, AWS, Go, GCP, Terraform. C++ shows up in a sixth, driven by silicon and performance work. TypeScript and React trail at a tenth or less, which matches the thin frontend headcount above.
Most of the hiring is engineering, but sales is already large
Software and applied-AI roles lead, no surprise. The bigger tell is the 1,317 go-to-market openings across the set. A field this early already staffing revenue teams tells you a lot of these products have paying customers.
The market is senior. Early-career barely registers
Across 9,714 roles, 261 are intern, new-grad, or residency. That is 2.7%. Staff, principal, and management make up far more. If you’re coming out of school, the wall is real, and the few doors that exist are concentrated in a handful of companies with structured new-grad programs (Databricks, Waymo, OpenAI, Anthropic, Sierra) rather than spread evenly.
In actual years, not labels
Titles only go so far, so the explicit “X+ years” line is pulled from every description that has one. About 45% state a number.
The median floor is 5 years and the single most common ask is 5+. Only 10% of these roles accept two years or fewer; 23% accept three or fewer. By specialty the bar is highest for forward-deployed and SRE work at 6 years, lowest for research at 4. Two caveats keep this from being as grim as it looks. The other ~55% of descriptions name no number and tend to be more flexible, and a “5+ years” line is often a filter rather than a hard gate, especially when the rest of the application is strong.
Which sectors are selling tells you which have customers
Sales as a share of each sector’s hiring is a rough read on how commercial it is. Security, data infrastructure, generative media, and applied enterprise are staffing revenue hard, a fifth to a third of their openings. Robotics, defense, and silicon sit near zero, still building rather than selling. Pick a sector partly on which stage you want to join.
San Francisco, then a steep drop
SF has more openings than the next several cities combined. Remote is second, ahead of any single city, but it’s still under a third of roles. New York and London are the only other real clusters. Location narrows the list fast.
The busiest boards
Databricks, OpenAI, and Anthropic each list hundreds of roles. Most are senior, so the volume doesn’t mean easy entry, but this is where the openings are.
The work has moved downstream
The hiring reads like an industry past its research phase. Only one technical role in ten builds or trains models; the rest sit downstream, scaling infrastructure, wiring up agents, and carrying models into a customer’s stack. Add the 1,317 open sales roles and these look like companies shipping products, not labs chasing a benchmark. The roles they list the most of are the least glamorous ones, and they lead into the parts of the business that already have paying customers.
Method: role data pulled live from the public Greenhouse, Lever, and Ashby job-board APIs across the 102 of 156 AI companies that expose one. Titles classified by keyword, so counts are approximate. Companies on Workday or custom boards aren’t included yet.