Most of what circulates about AI and jobs is speculation, some of it careful and well-researched, some of it not. The Stanford Canaries Dashboard is different: it is real, measured payroll data updated monthly, not a scenario or a forecast, and it is one of the few sources that can actually tell you whether AI-related job disruption is showing up yet, and if so, where.
The Stanford Canaries Dashboard was built by the Stanford Digital Economy Lab in collaboration with ADP Research, led by economist Erik Brynjolfsson along with researchers Bharat Chandar and Ruyu Chen. It draws on real payroll records from ADP, the largest payroll processing firm in the country, covering roughly 25,000 firms and 4.6 million workers across more than 730 distinct occupations.
This guide breaks down what the Stanford Canaries Dashboard actually measures, what it has found so far, and what it means for your own job search, especially if you are early in your career or working in a role with high exposure to AI tools.
Speculative scenarios like AI 2027 and AI 2040 describe what might happen. The Stanford Canaries Dashboard is one of the better tools for checking which parts of those scenarios are actually showing up in real employment data right now.
The Short Answer
The Stanford Canaries Dashboard tracks real ADP payroll data since ChatGPT’s release in November 2022 and finds that employment growth is slowest in the most AI-exposed occupations, with the clearest effect concentrated among early-career workers ages 22 to 25. Software developers in that age group show large declines, while less-exposed roles like home health aides show early-career gains instead.
The dashboard also finds a real gender gap in who is affected, driven mostly by women being more concentrated in AI-exposed occupations to begin with, rather than unequal treatment within the same job.
This is measured correlation in real payroll data, not a confirmed causal story, and the researchers themselves are explicit about that distinction.
What the Stanford Canaries Dashboard Actually Measures
| AI Exposure Level (Ages 22-25) | Women’s Annual Employment Growth | Men’s Annual Employment Growth |
|---|---|---|
| Most AI-exposed occupations | -4.5% per year | -2.5% per year |
| Least AI-exposed occupations | +1.3% per year | +2.7% per year |
The table above reflects the dashboard’s most striking finding: a real, measured gender gap in early-career employment growth that widens sharply as AI exposure increases. The methodology groups workers into five exposure groups based on established AI-exposure research, then tracks employment trends within firms that have a continuous five-year history in ADP’s system, rather than trying to capture new firm formation or firm exits.
Occupations are matched using the federal government’s standard occupation classification crosswalk, and workers are further broken down by age and gender to see whether disruption, where it exists, is concentrated in specific groups rather than spread evenly.
The researchers are careful to describe this as a correlation between occupational AI exposure and employment trends, not a proven causal link. They note that some earlier declines were likely driven by unrelated factors, and that a statistically significant AI-exposure effect only became clear starting in 2024 once broader controls were applied to the data.
What the Data Actually Shows About Entry-Level Jobs
Software developers are the clearest example in the dashboard. Early-career developers, ages 22 to 25, show large declines in employment since ChatGPT’s release, while slightly older developers show more modest declines, and developers past early-career age show continued growth. Customer service representatives show a similar pattern: noticeable declines for the youngest two age groups, with growth continuing for older workers in the same occupation.
Lower-exposure occupations tell a different story entirely. Stock clerks show little relationship between age and employment trends at all, and home health aides, one of the least AI-exposed occupations tracked, actually show early-career workers gaining more than other age groups in that same role.
This pattern, concentrated disruption in early-career workers within high-exposure occupations specifically, is the single most consistent signal the Stanford Canaries Dashboard has produced so far, and it lines up with some of the specific claims in speculative documents like AI 2027 about junior technical roles.
The Gender Gap the Dashboard Found
In the most AI-exposed occupation group, early-career women saw employment contract by 4.5 percent per year, compared to a 2.5 percent annual contraction for early-career men in the same exposure group. In the least-exposed group, women grew at 1.3 percent per year versus 2.7 percent for men, meaning women trailed men in both directions.
The researchers found this gap is driven mainly by occupational sorting rather than unequal treatment within identical jobs. Women make up 43.8 percent of workers in the most-exposed occupation category and 21.2 percent in the second-most-exposed category, compared to 32.4 percent and 18.1 percent for men in those same two categories respectively.
In plain terms, women are simply more concentrated in the specific occupations the dashboard identifies as most exposed to AI, which explains most, though not necessarily all, of the gap in outcomes between men and women shown in the data.
What This Means If You Are Job Searching Right Now
If you are early in your career in a high AI-exposure occupation, particularly software development, this is the strongest piece of real evidence available that entry-level hiring in your specific field is measurably softer than it was before late 2022. That is a meaningfully different claim than a speculative scenario, since it reflects payroll data from millions of actual workers rather than a narrative prediction.
The most useful response is not panic, since the declines described are real but not catastrophic, and not dismissal either, since the pattern has held consistently enough across multiple occupations to be more than noise. Building demonstrated skill working alongside AI tools, rather than only competing on tasks AI already handles well, remains one of the more concrete responses available regardless of your specific occupation or exposure level.
If your occupation shows up as lower-exposure, this data is genuinely reassuring rather than something to second-guess. The dashboard’s own findings suggest disruption so far is real but concentrated, not the broad, all-occupation disruption sometimes implied by more speculative sources.
Common Mistakes People Make When Interpreting This Data
- Treating correlation as proof of causation. The dashboard’s own researchers describe their findings as a correlation, not a confirmed causal link between AI adoption and job losses.
- Applying findings from one occupation to your own without checking exposure level. Software development shows large early-career declines, but many other occupations show little to no comparable effect.
- Ignoring the gender-gap explanation. Assuming unequal treatment within identical jobs, rather than occupational concentration, misreads what the dashboard’s own breakdown actually shows.
- Citing the dashboard once and never checking it again. It updates monthly, and a data source this current is far more useful as an ongoing check than a one-time read.
Any one of these mistakes is understandable given how unfamiliar most job seekers are with reading labor economics data directly, but together they explain why the same dashboard gets cited to support wildly different conclusions depending on who is citing it.
Frequently Asked Questions About the Stanford Canaries Dashboard
Is the Stanford Canaries Dashboard the same kind of document as AI 2027 or AI 2040?
No. AI 2027 and AI 2040 are speculative scenarios written by independent researchers describing possible futures. The Stanford Canaries Dashboard is real, measured payroll data from millions of actual workers, updated monthly, which makes it a fundamentally different and generally more reliable type of source for understanding what is happening right now.
Does the dashboard prove AI is causing job losses?
Not exactly. The researchers themselves describe their findings as a correlation between AI exposure and employment trends, not a confirmed causal relationship. Other factors likely contributed to earlier declines, with a clearer AI-specific signal only emerging in the data starting in 2024.
Which jobs does the Stanford Canaries Dashboard show as most affected?
Software development shows some of the clearest early-career declines in the dataset. Customer service representative roles show a similar, if less dramatic, pattern. Lower-exposure occupations like stock clerks and home health aides show little to no comparable decline for early-career workers.
Why do women show worse outcomes in this data?
Mainly because women are more heavily concentrated in the specific occupations the dashboard classifies as most AI-exposed, not because of unequal treatment within the same job. Correcting for occupational sorting substantially narrows, though may not fully eliminate, the gap the dashboard identifies.
Should this data change my career plans?
It is a stronger signal than a speculative scenario, since it reflects real, current data rather than a narrative prediction. It is still one input, not a complete picture, and should be weighed alongside your specific occupation, location, and experience level rather than treated as a universal rule.
The Bottom Line
The Stanford Canaries Dashboard is one of the most useful tools available right now for separating real, measured AI-related labor market effects from speculation. It shows a real, if concentrated, pattern: slower or declining early-career employment in high AI-exposure occupations, most clearly in software development, alongside a gender gap explained mostly by which occupations women and men are concentrated in.
None of this amounts to proof of a specific timeline like the one described in AI 2027, and the researchers are careful to say so themselves. But it is a genuinely useful, real-world check against speculative scenarios, and it is worth revisiting periodically since the dashboard updates monthly as new data comes in.
Anonymous and approximate. We store one number for this article and nothing about you.
Email digest
We also run a layoff tracker, built from company filings and official notices. A plain email summary of what changed on these trackers: headline numbers, the largest new entries, and links to the sources. You confirm your address by clicking a link we email you, and one click unsubscribes. Already subscribed? Use this form to change what you get. The boxes replace what you had.
Our mail provider records whether you open an email and which links you follow, including the confirmation email, which is measured before you have agreed to anything. Unsubscribing stops the sending and the recording together.
You can read every past edition first. Each one is kept permanently, exactly as it was sent.
Privacy note: what we store and how to erase it
What we store: your email address, the choices above, and timestamps (signed up, confirmed, last sent). Nothing else about you.
What our mail provider records: whether you opened an email and which links you followed, tied to your address. That is Brevo, the service that delivers these emails, and it works by adding a small invisible image and by routing links through its own address. We read it to see which sections people use. Unsubscribing stops the sending and the measuring at the same time, and erasing your row removes it from our side.
The confirmation email is measured too: the same invisible image is in the one message that asks your permission. If you open it, Brevo records that against your address before you have agreed to anything, and it records it even if you then never confirm. We would exempt that message if we could. Brevo offers no way to switch its measuring off for a single email, and the one setting that exists covers every email we send or none of them. If you would rather not be measured at all, do not open it. Nothing else is ever sent to an unconfirmed address, and the signup is deleted from our side after 30 days.
About the links: links in the digest pass through a counter on this site that adds 1 to a total for that link and sends you straight on. It records no identifier, no IP address and no browser details, so it counts how many times a link was followed and can never say who followed it.
What it is used for: sending you exactly the emails you ticked, nothing else. The address is never shared, sold, or used for any other purpose.
How to erase it: click the unsubscribe link in any email. That stops all sending immediately, and unsubscribed addresses, along with signups that were never confirmed, are hard-deleted automatically after 30 days. You can also ask via the contact page.