The week in AI hiring — Oct 21, 2026.

A fixed snapshot of the 693 AI-core roles open at 60 AI companies on Oct 21, 2026. Same method as the daily page, one run, no mixing of dates.

693

open AI-core roles in this snapshot

So what: every one is a live, validated opening from that day's run — not a rolling average.

AI companies tracked weekly

So what: the same 60 boards every week — the comparison is like for like.

16%

of roles are remote (111 of 693)

So what: the rest are tied to a place — check location before promising remote work.

21%

publish a salary range (146 of 693)

So what: those 146 are the roles where the budget is already approved.

Δ vs last week 6-day window

Oct 15, 2026 → Oct 21, 2026. Roles first seen, roles no longer listed, and the boards that moved most. Read from the same stable IDs that power the daily feed, so a repost is not counted as a new role.

roles opened — first seen in the window

roles closed — no longer listed

−11

net change in open roles (704 → 693)

Biggest increases

Anthropic45 → 48 open+3

Reddit34 → 37 open+3

Wayve29 → 31 open+2

Biggest decreases

Helsing16 → 8 open−8

Nebius27 → 22 open−5

GitLab13 → 10 open−3

Baseline note: Tracking history starts Oct 14, 2026 and full 60-board coverage starts Oct 15. This first edition compares against Oct 15 (6 days); comparing to Oct 14 would label 585 first-seen roles as new hires. Later editions use an exact 7-day window.

Salary signals

Only 146 of 693 roles (21%) publish a range at all. The companies that do are setting the market — and the top of it moved this week.

OpenAI

Research Scientist / Research Engineer - Foundations, Search — highest published range at this company

$380k$850k

SCALE 0 — $900K

Wayve

Principal Research Scientist, Robot Foundation Models — highest published range at this company

$370k$513k

SCALE 0 — $900K

Widest bands in this snapshot

OpenAI $380k – $850k $470k wide 57 roles with pay

Mercor $160k – $500k $340k wide 6 roles with pay

Mercor $180k – $500k $320k wide 6 roles with pay

So what: the wider the band, the more room the budget has for the right candidate.

5 companies publish a range that reaches $400k or more this week. So what: a published range means an approved budget — start that conversation at the number, not at "what's the comp?".

Hiring volume leaderboard

Open AI-core roles per company, Oct 21, 2026 snapshot. The interesting story is in how each one hires, not just how much.

01OpenAI

81 roles 1 of 81 remote · seniority not stated in titles · asks for Agentic systems, Reinforcement learning, Spark. So what: 1% remote — check location before you promise remote work.

02Waymo

77 roles 0 of 77 remote · 20 staff, 19 senior, 17 intern · asks for PyTorch, JAX, TensorFlow. So what: Onsite-only: the candidate pool is local, so whoever owns the local pipeline owns the fill.

03Anthropic

48 roles 4 of 48 remote · 3 staff, 0 senior · asks for GPT, Agentic systems, Reinforcement learning. So what: 8% remote — check location before you promise remote work.

04Reddit

37 roles 37 of 37 remote · 21 staff, 9 senior · asks for PyTorch, TensorFlow, Spark. So what: All remote — the easiest placement profile to close: no relocation, no commute.

05Wayve

31 roles 0 of 31 remote · 8 staff, 9 senior · asks for PyTorch, Spark, CUDA. So what: Onsite-only: the candidate pool is local, so whoever owns the local pipeline owns the fill.

06Mistral AI

31 roles 0 of 31 remote · 1 staff, 1 senior, 3 intern · asks for PyTorch, RAG, Fine-tuning. So what: Onsite-only: the candidate pool is local, so whoever owns the local pipeline owns the fill.

07Anduril

29 roles 1 of 29 remote · 2 staff, 11 senior · asks for PyTorch, TensorRT, TensorFlow. So what: 3% remote — check location before you promise remote work.

08Databricks

23 roles 0 of 23 remote · 9 staff, 7 senior, 1 intern · asks for Databricks, Spark, Fine-tuning. So what: Onsite-only: the candidate pool is local, so whoever owns the local pipeline owns the fill.

09Scale AI

22 roles 3 of 22 remote · 7 staff, 3 senior, 1 intern · asks for Agentic systems, Fine-tuning, Reinforcement learning. So what: 14% remote — check location before you promise remote work.

10Nebius

22 roles 8 of 22 remote · 1 staff, 20 senior · asks for TensorRT, CUDA, Distillation. So what: 36% remote — check location before you promise remote work.

Stuck Hiring Index

The roles companies can't close — and what that says about the gap between the talent they want and the talent that exists.

Warming up

The first ranking lands mid-November

A role only counts as "stuck" with real history behind it — how many days it's been open, and whether the company quietly reposted it hoping nobody noticed. Our daily history started accumulating on Oct 15, 2026, and a fair ranking needs 35 days of it. Rather than guess, we're waiting for the data.

Already working: content fingerprinting links the same role re-appearing under a new listing to its original, so a repost cannot reset the clock.

Days openEach posting is tracked daily by a stable ID, so we see its true age — not the date on the latest repost.

Repost detectionContent fingerprinting catches the same role reappearing under a fresh listing and links it to the original.

The rankingCompanies ranked by how long their hardest roles stay open. Long-lived + reposted = struggling to hire.

Subscribe above to get the inaugural Stuck Hiring Index the day it drops.

Skills snapshot

What 693 job descriptions actually ask for this week. Two skills lead the list.

PyTorchthe default framework — assumed, not asked

242 / 693 · 35%

Agentic systemsthe agent stack is now a line item

171 / 693 · 25%

Also in demand at the top of the board: PyTorch, JAX, TensorFlow (Waymo) · GPT, Agentic systems, Reinforcement learning (Anthropic).

153

staff-level roles open. So what: companies are buying experience; lead with senior candidates.

124

senior roles open — staff + senior = 40% of all demand. So what: that's where recruiting time earns its keep.

111

remote roles (16%) — the split is company-by-company, not a trend line. So what: pitch the company, not the average.

Where the roles are — and who can take them

Geography and visa policy from the Oct 21, 2026 edition — 693 roles, schema 1.6.0.

01United States

489 roles · 71%

02United Kingdom

68 roles · 10%

03Canada

36 roles · 5%

04France

25 roles · 4%

05Germany

20 roles · 3%

97%

of roles carry a country tag — 674 of 693. Location is stated in the posting, not inferred.

193

roles state a visa policy at all. The other 500 are silent on it.

121

of those 193 — 63% — explicitly offer visa sponsorship.

Methodology & data source

Where this edition's numbers come from

Every figure on this page is read from one run — the daily run of Oct 21, 2026 — over the public applicant-tracking boards (Greenhouse, Lever, Ashby, Workable) of 60 AI companies. The run produced 693 validated AI-core rows, exported to CSV and counted from that export, not re-derived from a different source.

Δ vs last week uses the same stable IDs as the daily feed, read from the public lifecycle map as it stood on Oct 21, 2026. A role counts as opened when it was first observed inside the window, closed when it was last observed before the snapshot. Reposts are linked to the original role, so they are not counted as new. map sha256 21f83571b241a96c

Remote means the role's workplace_type as reported by the company's own ATS. 279 of 693 roles (40%) state no workplace policy, so they are counted as neither remote nor onsite — the 16% remote figure is a floor, not a ceiling.

The underlying dataset is public: the same feed powers the AI/ML Jobs Scraper on Apify, where anyone can run it, inspect a run, or pull the raw rows and check the arithmetic.

Edition: Week of Oct 21, 2026
Data snapshot: Oct 21, 2026
Published: Oct 22, 2026
Schema version: 1.6.0
Roles in snapshot: 693 AI-core
Companies tracked: 60
Source export: ai-jobs-2026-10-21_run-fhnkS9D2SEtxEgVVG.csv