Capital One Is the Only Company in Our Index Whose AI Hiring Is Actually Shrinking
Across the 50 companies HireIndex tracks week over week, exactly one has posted fewer AI roles in every measurement period than the period before it. It isn’t a struggling startup. It’s Capital One, which four weeks ago had more open AI roles than Scale AI, Amazon, or Reddit.
The August 3, 2026 snapshot covers 1,623 open AI/ML roles across 765 companies. Capital One’s count over the four tracked weeks reads 49, 49, 45, 36 — a 26.5% drop, and the only monotonic decline anywhere in the index.
How many tracked companies are actually shrinking?
Ten of fifty, counting any net loss across the four weeks. Twenty-eight grew and twelve finished flat. Net movement across the tracked set was positive, at 37 roles added.
| Company | Jul 13 | Aug 3 | Change |
|---|---|---|---|
| Mercor | 28 | 6 | −22 |
| Capital One | 49 | 36 | −13 |
| J.P. Morgan | 16 | 9 | −7 |
| Plaid | 12 | 9 | −3 |
| Commonwealth Bank of Australia | 9 | 6 | −3 |
| 11 | 9 | −2 | |
| Royal Bank of Canada | 10 | 8 | −2 |
| JPMorganChase | 7 | 5 | −2 |
| Autodesk | 5 | 4 | −1 |
| Labelbox | 4 | 3 | −1 |
Most of those are small absolute moves on small bases. Capital One is the outlier — a large employer shedding a quarter of its posted AI headcount in a month, in a straight line.
Why does Capital One still show an accelerating signal?
Its signal label reads “accelerating” and its score sits at 10, the top of both scales. Five of the ten declining companies carry that same label.
The label isn’t measuring direction. It’s derived from what a company posts rather than how much: seniority mix, breadth across AI functions, whether the roles are platform work or one-off backfills. By that measure Capital One still reads as serious. The 36 remaining roles are Senior Lead AI Engineer, Lead AI Engineer, Applied Researcher, and Lead Machine Learning Engineer, covering LLM inference, agentic infrastructure, and AI compute. The seniority split is 17 mid-level, 15 senior, and 4 senior-heavy, with no entry-level seats at all.
A company can be building something ambitious and hiring less for it in the same quarter. The label captures the first condition and the trend line captures the second, so read both.
What overtook Capital One at the top?
Scale AI, moving the opposite direction over the same four weeks: 45, 47, 52, 59. That’s a 31.1% rise and the only clean monotonic climb among the large employers.
| Company | Open roles (Aug 3) | Four-week trend |
|---|---|---|
| OpenAI | 92 | 90 → 85 → 93 → 92 |
| Databricks | 69 | 69 → 65 → 69 → 69 |
| Scale AI | 59 | 45 → 47 → 52 → 59 |
| Amazon | 50 | 47 → 51 → 56 → 50 |
| 39 | 37 → 39 → 40 → 39 | |
| Capital One | 36 | 49 → 49 → 45 → 36 |
On July 13, Capital One sat above Scale AI by 49 roles to 45. Four weeks later Scale AI is ahead by 23.
The two are hiring for different work. All 59 Scale AI roles are remote, while Capital One’s 36 split into 18 remote, 17 in New York, and 1 in Toronto. Scale AI’s mix runs 39 mid-level, 11 senior-heavy, 5 senior, 3 leadership, and 1 junior — that single seat being the only entry-level role between the two companies.
Which declines are real and which are scraper noise?
Mercor shows the steepest percentage drop in the index, falling from 28 to 6. Treat that figure with suspicion. Its four weeks read 28, 3, 31, 6: the count collapsed, recovered past its starting point, then collapsed again. No company hires and unhires 25 people twice in a month. That pattern points to a feed problem, where a job board intermittently fails to return the full listing set.
Three other companies show the same shape. OpenAI reads 90, 85, 93, 92 — an 8-role swing around what is effectively a flat line. Royal Bank of Canada runs 10, 5, 3, 8. Barclays runs 4, 0, 0, 6.
The test is whether weekly swings are larger than the net change. A company that moves 25 roles week to week and finishes 22 down is producing noise with a coincidental endpoint. A company that moves 4 and then 9 in the same direction and finishes 13 down, as Capital One did, has done something real.
This matters for anyone reading hiring data to time an application. A single week-over-week delta tells you almost nothing. Direction held across three consecutive measurements is the shortest interval worth acting on, which is why four weeks is the minimum window we publish.
What does this mean if you’re targeting Capital One?
Fewer seats at the same bar. The remaining roles are Lead and Senior Lead titles, which map to senior and principal levels elsewhere. LLM Engineer work there concentrates in customization, fine-tuning, and gateway infrastructure, while the AI Platform postings cover foundation model hosting and inference.
Candidates who were treating Capital One as the surprise volume employer in enterprise AI — it outposted every frontier lab on Machine Learning Engineer titles earlier this year — should recalibrate against Scale AI and Databricks, which now carry 59 and 69 open roles respectively.
Capital One’s 36 remaining roles are split 18 remote and 17 in New York, with a single seat in Toronto.