This note captures the state of AI engineering hiring on 2026-08-24, pulled directly from the AI Dev Jobs public API. The numbers are not a survey. They are a live, daily-refreshed index of what companies are actually posting to their own applicant tracking systems right now — scraped continuously from the AI Dev Jobs ATS source feed network, deduplicated, and canonicalized.
Top 10 hiring companies right now
The concentration at the top is striking. OpenAI, LILT, and Anthropic alone account for 1003 open roles — roughly 11% of the entire index. The top 10 companies account for 2,081 roles, or 23.2% of the market. This is a market with a long tail (476 companies below the top 10) but also with serious pockets of single-company acceleration.
| Company | Open roles | Avg salary |
|---|---|---|
| OpenAI | 417 | $322,091 |
| LILT | 307 | — |
| Anthropic | 279 | $399,738 |
| Anduril | 203 | $198,018 |
| Graphcore | 176 | $283,600 |
| Nebius | 163 | $177,839 |
| Scale AI | 148 | $234,281 |
| FluidStack | 144 | — |
| Harvey AI | 134 | $225,655 |
| Waymo | 110 | $256,112 |
The frontier labs (OpenAI, Anthropic, xAI) pay a premium of roughly $131k over the defense-tech, autonomy, and infrastructure players in the same leaderboard. That gap is the clearest signal in the data about where investor capital is being deployed most aggressively right now.
Top demanded skills
LLM work now dominates the index. 2,611 of 8,967 roles (29.1%) list llm as a tag. agents is close behind at 2,770 (30.9%), and generative-ai sits at 1,685 (18.8%). A year ago pytorch and deep-learning led by volume. The demand center of gravity has migrated up the stack — from model training to model orchestration and agent design.
| Tag | Role count | Avg salary |
|---|---|---|
| llm | 2,611 | $249,735 |
| agents | 2,770 | $233,394 |
| generative-ai | 1,685 | $242,425 |
| distributed-systems | 1,459 | $268,306 |
| pytorch | 953 | $256,903 |
| fine-tuning | 860 | $252,160 |
| research | 748 | $298,827 |
| reinforcement-learning | 532 | $288,466 |
| mlops | 504 | $230,688 |
| gpu | 535 | $248,223 |
Research roles command the highest average salary ($298,827) among tags with 500+ roles, followed by reinforcement learning and search. The premium for specialized, harder-to-hire skills is intact — training infrastructure and eval/reliability work (distributed systems, MLOps, GPU) continues to outpay generic application work.
Salary distribution
Of the 3,480 roles that publish salary ranges, the shape is bimodal around the $200k line. The $200-250k band is the single largest bucket (1,037 roles, 29.8%), with $150-200k close behind (911 roles, 26.2%). Everything below $150k is a minority (363 roles combined, 10.4%), and roles above $300k are a meaningful but not overwhelming slice (593 roles, 17.0%).
| Range | Roles | Share |
|---|---|---|
| Under $100k | 72 | 2.1% |
| $100k-$150k | 291 | 8.4% |
| $150k-$200k | 911 | 26.2% |
| $200k-$250k | 1,037 | 29.8% |
| $250k-$300k | 576 | 16.6% |
| $300k-$400k | 424 | 12.2% |
| $400k+ | 169 | 4.9% |
Workplace mix
Onsite is still the largest category by volume (4,793 roles, 53.5%), but hybrid roles pay the highest on average: $274,679 versus $220,924 for onsite and $224,682 for remote. The $49k hybrid premium is real and worth pausing on — it suggests the companies paying the most for senior talent right now want people in the building at least part of the week. Remote pay tracks onsite almost exactly.
| Workplace | Roles | Share | Avg salary |
|---|---|---|---|
| Onsite | 4,793 | 53.5% | $220,924 |
| Remote | 2,595 | 28.9% | $224,682 |
| Hybrid | 1,579 | 17.6% | $274,679 |
The ecosystem side
Hiring demand is not the only signal. On the infrastructure side, NothingHumanSearch — an independent index of agent-ready web services — now tracks 5,197 sites with agent discovery files (llms.txt, OpenAPI, ai-plugin), of which 673 have a live-verified MCP server over JSON-RPC, and 3,036 publish an llms.txt. Developer tools (1,578 sites) and AI-native tools (1,112 sites) are the two largest categories. Read alongside the hiring data, these two indexes describe the same market from opposite ends: 8,967 humans being hired to build AI products, into a world where 5,197 services are already exposing themselves natively to AI agents.
The story the data tells: the stack is diversifying faster than headcount is. Agent frameworks, eval pipelines, MCP servers, vector infra, and MLOps tooling are all real sub-markets now. Companies that want to hire into this market need to be specific about which layer they are hiring for — generic "ML engineer" listings are competing against a labor pool that self-identifies by framework and problem domain.
Methodology
Data in this note was pulled live at publication. The aidevboard.com index scrapes applicant-tracking feeds (Ashby, Greenhouse, Lever, Workday, custom careers pages) on a daily cron, canonicalizes titles and tags with a rules-based classifier, and dedupes by (company, title, location). The /api/v1/stats endpoint is public and unauthenticated. NHS data is from /digest.json — an index that live-probes sites for agent-discovery signals and MCP endpoints. Both APIs are agent-readable. This page auto-regenerates weekly.
Download raw data: The top-hiring-companies leaderboard is mirrored as a public gist — CSV · Markdown · view on GitHub. Auto-updated every weekly regeneration, canonical raw URLs are stable across revisions.
What's next
For the organizational implications of this hiring mix — specifically why the agents tag growing 30.9% of the index matters more than the raw salary numbers — see The Agentic Accountability Gap and Beyond the Prompt. For what those 6% of companies actually capturing returns are doing differently, see The Six Percent. Full reading paths at the Research Atlas.