# What does the 2025–2026 data really show about AI and jobs?

Date: 2026-10-28 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/ai-and-jobs-data-2026/

**Answer:** Across the whole U.S. labor market, the data through the summer of 2026 still show no clear wave of AI-driven job losses. The clearest signal is narrower: in the occupations most exposed to AI, employers are hiring fewer workers aged 22 to 25, especially where AI does the task rather than helping a person do it.

Headlines about AI and jobs swing between "nothing is happening" and "the white-collar collapse has begun." Both can quote a study. The way through is to look at what each measure can and cannot see, because in 2026 two careful research teams are watching the same economy and finding different things at different scales.

This article updates the site's earlier pieces on [entry-level jobs](/en/learn/ai-changing-entry-level-jobs/) and [growing occupations](/en/learn/jobs-growing-because-ai/) with data through mid-2026.

## The wide view: no clear disruption yet

The Budget Lab at Yale tracks the whole U.S. labor market every month using the Current Population Survey, the government household survey behind the unemployment rate. Its September 2026 update, using August data, finds no clear evidence of disruption tied to AI. The mix of occupations is shifting only slightly faster than it did when the internet spread in the late 1990s, and that shift began before ChatGPT. Workers in the jobs most exposed to AI are not becoming unemployed at higher rates, and a statistical comparison of exposed and unexposed workers shows no clear AI footprint so far.

The researchers are careful about what this means. Their measures of "exposure" are proxies built partly from AI companies' own data, and past technologies took decades, not months, to reshape work. Absence of a signal today is not a forecast.

## The close view: fewer young hires in exposed jobs

Stanford economists Erik Brynjolfsson, Bharat Chandar and Ruyu Chen use a different lens: payroll records from ADP, a large payroll company, covering millions of workers. Their August 2026 update, with data through June 2026, also finds no economy-wide displacement. But inside the most AI-exposed occupations, employment of workers aged 22 to 25 is 19% lower than it would have been if it had grown at the same pace as for young people in less-exposed jobs. Experienced workers in the same jobs show no such gap, and the gap has widened since the authors first reported it in August 2025.

Three details sharpen the picture. The decline comes mostly from companies hiring fewer young people, not from layoffs. It is concentrated where AI use mostly **substitutes** for a task (the AI does it) rather than **complements** a worker (the AI helps a person do it); in complementary uses, employment is flat or rising. And pay has not moved much; the adjustment is in the number of jobs. The authors call these early warning signs, "canaries in the coal mine," not proof of cause, and note that some trends began before generative AI.

## Why both can be right

The two findings fit together. A drop in hiring for one age group in a set of occupations is too small to move national totals, but it matters a great deal to the graduates trying to enter those fields. The U.S. Bureau of Labor Statistics took a similar view in its 2025 projections: it expects slower growth or decline where core tasks are easy for generative AI to copy, such as medical transcription and customer service, while still projecting growth for software developers because cheaper software may raise demand.

## A worked example: auditing your own tasks

The Stanford distinction between substituting and complementing gives you a practical way to look at your own job. List your weekly tasks and sort them.

| Task | Share of my week | AI's role today | Exposure |
|---|---|---|---|
| Drafting routine client emails | 20% | Can write a full draft | High: substitution |
| Summarizing meeting notes | 10% | Can produce the summary | High: substitution |
| Analyzing a budget with a manager | 15% | Speeds up the spreadsheet work | Medium: complement |
| Resolving a client's complaint by phone | 25% | Little | Low |
| Training a new colleague | 10% | Can prepare material | Low: complement |

You can ask an assistant to help you build the first draft of this table:

```prompt
Here is a list of the tasks in my job as an office administrator, with the rough share of my week each takes: [paste list].
For each task, tell me whether current AI tools could mostly do it, mostly help me do it, or barely help.
Explain your reasoning in one line per task and flag where you are unsure.
```

Treat its answer as a starting point; you know which tasks hide judgment, relationships or responsibility that a list does not show.

> **Try it:** Write down ten tasks from your last two weeks of work, estimate the share of time each took, and mark each one "AI does it," "AI helps me" or "AI barely helps." Add up the share in the first column.

## Which tasks in your work are most exposed?

The most exposed tasks share three features: they start and end on a screen, they follow a pattern that can be described, and their result can be checked quickly. Drafting standard documents, summarizing, transcribing, answering routine questions, and first-pass coding all fit. This is why customer service and entry-level software roles appear in both the BLS and Stanford analyses.

Tasks that resist are the ones that depend on trust, physical presence, judgment under uncertainty, or accountability for a decision: negotiating, caring for people, handling exceptions, and signing off on work. Notice too that exposure at the task level is not the same as risk to a job. If AI takes over a fifth of your week, the question becomes what you do with the freed time.

If the first column of your audit is large, the data suggest two moves. Shift your time toward the "AI helps me" column by learning to direct and check AI output well, since that is where the Stanford data show employment holding up. And build the experience that young workers now find harder to get: handling the exceptions and the people, which is the part of the job that a pattern cannot describe.

## Sources

- Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence: https://digitaleconomy.stanford.edu/wp-content/uploads/2025/08/Canaries_BrynjolfssonChandarChen.pdf
- Tracking the Impact of AI on the Labor Market: https://budgetlab.yale.edu/research/tracking-impact-ai-labor-market
- Incorporating AI impacts in BLS employment projections: occupational case studies: https://www.bls.gov/opub/mlr/2025/article/incorporating-ai-impacts-in-bls-employment-projections.htm