AI exposure and hiring demand
Disruption Map
This map places 387 US occupations by how much of their work is already done with AI at a desk, against whether employers want more people or fewer than a year ago. A separate view shows where physical AI could land first. It reads occupations, not people: where you go inside yours depends on the work you choose to own.
Four zones
The zones use cognitive AI exposure only: the share of work time on tasks where people already use AI for work. It counts as high at 20% of work time or more, which 93 occupations reach. Physical work share is shown as a separate view, because it measures how physical a job is, not whether robots can do it today.
AI-boosted
51 of 387 US occupations
High cognitive AI exposure, and hiring is rising or steady.
Much of this work is desk work people already hand to AI, and employers still want more people. A market like this rewards the people who use the new tools to take on bigger problems and who own the judgment and accountability around the result.
Next step. Use the demand while it lasts. Go after the bigger role, and build it around the blind spot of AI.
- Which high-paying jobs have rising demand?
- Is now a good time to ask for a raise or promotion?
- Which skills are employers paying a premium for?
Your next move up from any occupation · Career strategy session, US$495
Disruption underway
42 of 387 US occupations
High cognitive AI exposure, and hiring is cooling or falling.
Much of this work is desk work people already hand to AI, and employers want fewer people than a year ago. The data cannot prove AI caused the cooling. It does show where the pressure is building.
Next step. Plan your next move while you still have several to choose from, and aim it at the blind spot of AI: judgment, accountability, trust and physical-world integration.
- How do I know if my role is about to be cut?
- Will AI take my job?
- My employer won't say what AI means for my role. What can I check myself?
Your next move up from any occupation · Career strategy session, US$495
Solid ground
183 of 387 US occupations
Lower cognitive AI exposure, and hiring is rising or steady.
Less of this work is the kind people hand to AI today, and employers still want people. That is a strong base to build from. Exposure can rise as AI improves, so come back to this reading when the measures update.
Next step. Negotiate from the demand you have, and keep widening the part of the work only you can do.
- Which jobs are AI-proof?
- How do I know if I'm offering myself at a discount?
- Which skilled jobs are growing fastest?
Your next move up from any occupation · Career strategy session, US$495
Other forces at work
111 of 387 US occupations
Lower cognitive AI exposure, yet hiring is cooling or falling.
AI is a smaller part of this story. Something else, such as a sector's own budget cycle, may be doing more of the work, and this data cannot say what.
Next step. Read the sector trend on your occupation page, and widen your options before the market narrows them.
- Which well-paid jobs are losing hiring demand?
- Why would job postings rise while the long-term outlook falls?
- Is the job market getting worse?
Your next move up from any occupation · Career strategy session, US$495
30 US occupations in the index have no cognitive AI exposure figure, so they are not placed. Counts use JDI ratings with data to Nov 13, 2026.
The map
Each dot is one US occupation. Across: cognitive AI exposure, or in the physical view, physical work share. Up: US hiring demand, from Falling at the bottom to Rising at the top. Colour shows the zone, or in the physical view, the potential group.
Robots deployed at scale today cover a small share of this work; this shows where physical AI would land first if deployment speeds up.
The map needs JavaScript. Every occupation and its figures are listed under Every occupation.
In the cognitive view, dots right of the dashed line have high cognitive AI exposure and dots above the solid line have hiring rising or steady, so each quarter is one zone. The physical view shows a potential, not a zone: 48 occupations have a high physical work share with hiring rising or steady, and 31 with hiring cooling. Within each demand row, dots are spread up and down only so they do not overlap.
Every occupation, by zone
Cognitive exposure is the share of work time on tasks where AI was already used for work. Physical work share is JDI's heuristic v1 score from 0 to 100, a potential that does not set the zone. Each row links to the occupation page and to its moves up.
AI-boosted 51 occupations
Disruption underway 42 occupations
Solid ground 183 occupations
Other forces at work 111 occupations
Not placed (no cognitive AI exposure figure) 10 occupations
| Occupation | Cognitive AI exposure | Physical work share (v1, potential) | US hiring demand | Next step |
|---|---|---|---|---|
| Buyers and purchasing agents | n/a | 14.3 | Steady | Moves up |
| Calibration technologists and technicians | n/a | 42.9 | Rising | Moves up |
| Clinical laboratory technologists and technicians | n/a | 45.3 | Cooling | Moves up |
| Emergency medicine physicians | n/a | 28.8 | Steady | Moves up |
| Entertainment and recreation managers, except gambling | n/a | 22.8 | Steady | Moves up |
| Fundraising managers | n/a | 12.5 | Steady | Moves up |
| Medical dosimetrists | n/a | 29.4 | Cooling | Moves up |
| Project management specialists | n/a | 15.9 | Rising | Moves up |
| Property appraisers and assessors | n/a | 19.3 | Cooling | Moves up |
| Substitute teachers, short-term | n/a | 19.3 | Cooling | Moves up |
In Canada
Both exposure measures are built from US occupation data, so this map is US-based. Canadian occupation pages show the reading for the matched US occupation, labelled as US-based. The index gives no national Canadian verdict, for the same reason it gives no national Canadian demand rating: the official Canadian outlooks are published province by province.
Disruption Watch
Product moves that put established companies on the back foot, as product facts. Share prices and executive news are left out.
- Product
Anthropic adds spreadsheets, whiteboards and meeting notes to Claude
Claude Sheets builds spreadsheets inside a Claude chat, with tabs, formulas, formatting and drop-downs, and exports to .xlsx or CSV. Claude Whiteboards is a shared planning board. Both are in beta on paid plans. Claude Meetings joins video calls, takes notes and updates documents, working with Google Meet at launch. Claude Data, formerly Claude Dashboards, turns raw figures into visual reports and is generally available on paid plans. Anthropic's help page lists limits for Sheets, including 100,000 cells and no pivot tables or macros.
Incumbent on the back foot. Microsoft Office and Google Workspace.
What it means for work. If the working day starts in an AI assistant rather than in Excel or Docs, the routine part of building a sheet or a report moves to the assistant. What stays with you is deciding what the numbers should answer, and standing behind them.
Source: Windows Latest, Nov 11, 2026, citing Bloomberg and TechCrunch, Nov 10, 2026.
- Product
OpenAI launches ChatGPT Atlas, a web browser with ChatGPT built in
Atlas launched on macOS, with Windows and mobile versions promised later. Its agent mode, which can carry out tasks on web pages, needs a Plus, Pro or Business subscription.
Incumbent on the back foot. Google Chrome.
What it means for work. The browser is where much office research and form-filling happens. An agent inside it can do the clicking; the judgment about what to look for and what to trust stays with the person.
Source: SiliconANGLE, Oct 21, 2025.
- Product
Claude starts creating and editing Excel, Word, PowerPoint and PDF files
Anthropic released a preview in which Claude creates and edits spreadsheets, documents, slide decks and PDFs, first for Max, Team and Enterprise plans, with Pro to follow.
Incumbent on the back foot. Microsoft Office.
What it means for work. The first draft of a model, memo or deck is becoming the AI's job. Review, accountability for the result and the conversation with the decision maker are not.
Source: TechRepublic, Sep 9, 2025.
Opportunity Watch
Disruption also creates openings. Each piece connects a dated change to occupations in the index and is labelled for what it is: a hypothesis, checked against JDI's data at every refresh.
Hypothesis
The data-centre power buildout and the people who wire it
What is happening. Alphabet, Amazon and Meta together expect to spend about $550 billion to $570 billion on capital investment in 2026, by JDI's sum of their own guidance, and Amazon's CEO said most of Amazon's share is for AI data centres (the figures and sources are on the AI Frontier page). Data centres need grid connections, substations and electrical work.
The hypothesis. If the buildout continues, demand should stay steady or rise for the people who design, build and maintain power systems.
What JDI's data shows. So far 4 of 4 are rising or steady in JDI's data. The data is consistent with the hypothesis. It does not prove it.
- Electricians: US hiring demand Rising (slightly rising), Solid ground
- Electrical engineers: US hiring demand Rising (rising), Solid ground
- Electrical and electronics repairers, powerhouse, substation, and relay: US hiring demand Rising (slightly rising), Solid ground
- Electrical and electronic engineering technologists and technicians: US hiring demand Rising (slightly rising), Solid ground
What would change our mind. Postings in these occupations' sectors turning down while capital spending stays high would weaken it.
Sources: Company earnings statements, Oct 2026, as recorded on the AI Frontier page; JDI ratings.
Hypothesis
More robots on factory floors, and the people who keep them running
What is happening. US factories installed almost 38,500 industrial robots in 2025, up 12% on 2024, according to the International Federation of Robotics.
The hypothesis. More machines on the floor should mean steady demand for the engineers and technicians who specify, install, program and repair them.
What JDI's data shows. So far 4 of 4 are rising or steady in JDI's data. The data is consistent with the hypothesis. It does not prove it.
- Mechanical engineers: US hiring demand Rising (rising), Solid ground
- Electro-mechanical and mechatronics technologists and technicians: US hiring demand Rising (slightly rising), Solid ground
- Industrial engineers: US hiring demand Rising (rising), Solid ground
- Engineers, all other: US hiring demand Rising (slightly rising), Solid ground
What would change our mind. Robot installations rising while these occupations cool would weaken it.
Sources: International Federation of Robotics, World Robotics 2026, Sep 24, 2026; JDI ratings.
Hypothesis
AI written into everyday work, and the people who secure and run it
What is happening. Indeed Hiring Lab counts 7.6% of US job postings mentioning AI or machine learning in Oct 2026, against 3.9% a year earlier. Each AI tool an organization adopts adds systems to secure, data to protect and decisions about where the tool can be trusted.
The hypothesis. As AI spreads through ordinary work, demand should stay steady or rise for the people accountable for securing and running the systems it depends on.
What JDI's data shows. So far 3 of 3 are rising or steady in JDI's data. The data is consistent with the hypothesis. It does not prove it.
- Information security analysts: US hiring demand Rising (slightly rising), AI-boosted
- Computer and information systems managers: US hiring demand Rising (rising), Solid ground
- Computer and information research scientists: US hiring demand Rising (rising), AI-boosted
What would change our mind. AI mentions in postings rising while these occupations cool would weaken it.
Sources: Indeed Hiring Lab AI tracker (CC BY 4.0); JDI ratings.
Method and sources
Version. Disruption Map v1, Nov 17, 2026. Physical AI exposure is heuristic v1. Each change to a measure, a threshold or a zone rule gets a new version number and a line in the changelog, so earlier readings stay comparable.
Cognitive AI exposure: the primary measure
The measure is Anthropic Economic Index observed exposure: the estimated share of an occupation's work time spent on tasks where people were already using Claude for work, from samples in August and November 2025, published March 5, 2026. Automated use counts in full and assisted use at half weight. The data is released under CC-BY. It covers 387 of the 417 US occupations in the index.
We chose it over the alternatives. It measures use at work, not what a model might do in theory, and it is the most recent of the published measures. It also counts only tasks that the 2023 "GPTs are GPTs" study judged feasible for a language model, so it builds on that work. Its limit is that it counts one AI assistant, so it understates AI use overall.
For reference we also matched the "GPTs are GPTs" theoretical exposure (Eloundou, Manning, Mishkin and Rock, 2023; human-rated beta, MIT licence) to 400 occupations. Across occupations the two measures correlate at 0.51: related, and far from the same. The reference measure is in the data file and is not used for zones. The AI Occupational Exposure index of Felten, Raj and Seamans (2021) is widely cited, but its data repository publishes no licence, so we do not use it.
Physical AI exposure (physical work share): heuristic v1, a potential
There is no public, occupation-level measure of exposure to robots and automated machines that we can use, so JDI built one from O*NET 31.0 (USDOL/ETA, CC BY 4.0), and labels it a heuristic. Each O*NET rating runs from 1 to 5 and is rescaled to 0 to 1.
- Handling = average of the importance of Handling and Moving Objects (Work Activities) and time spent using your hands to handle, control or feel objects, tools or controls (Work Context).
- Operating = average of the importance of Controlling Machines and Processes and of Operating Vehicles, Mechanized Devices, or Equipment (Work Activities).
- Physical share = average of handling and operating.
- Routine = average of time spent making repetitive motions, the importance of repeating the same tasks, and how far the pace is set by the speed of equipment (Work Context).
- Physical AI exposure = 100 × physical share × (0.5 + 0.5 × routine).
The routine factor scales the physical share between half and full weight, because robots are deployed first in repetitive, machine-paced work. Rows O*NET flags for suppression are dropped. Where several O*NET occupations map to one occupation in the index they are averaged. It covers 391 of the 417 US occupations. Across occupations it correlates at -0.40 with cognitive exposure, so the two measures mostly pick out different work.
Why it does not set the zones. Robots deployed at scale today cover a small share of this work; this shows where physical AI would land first if deployment speeds up. Folding it into the zones would read a high physical share as disruption, or as a boost, when robots do little of that work today. So the zones use cognitive exposure alone, and the physical score has its own view, with its own labels: physical work with hiring rising, physical work with hiring cooling, and less physical work. Occupation pages show it as one line, "Physical work share", with no zone.
Limits of v1. It measures how much of the work is the kind of physical activity robots are being built for. It does not measure whether a robot can do that work today. Skilled trades that work in varied, unpredictable settings, such as electricians or mechanics, score high, although robots handle unstructured sites poorly; a version 2 candidate adds a weight for how structured the work setting is. O*NET ratings come from surveys in different years, and the score says nothing about the cost of a robot against a wage. Read it as direction. It is labelled as a heuristic and will be refined.
Bands and zones
Cognitive exposure is low under 5% of work time, moderate from 5% to 20%, and high at 20% or more. Physical exposure is low under 20, moderate from 20 to 40, and high at 40 or more. Zones use the cognitive measure only; an occupation with no cognitive figure is not placed. In the physical view, a share of 40 or more counts as physical work. On the hiring side, the demand states Rising and Steady count as rising or steady, Cooling and Falling as cooling. Mixed ratings, where postings and the official forecast disagree, follow the postings signal, the more recent of the two. Zones are recomputed at every data refresh.
What the map cannot tell you. Whether AI caused a change in hiring. Your own risk, which depends on your employer, your role and your record. How exposure differs by level of seniority. These are the same limits every JDI reading carries.
Data files: exposure.json, zones.json. Built Nov 19, 2026. Physical AI milestones are tracked on Physical AI. How the hiring ratings are built: method.
Sources
These links open in a new tab.
- Anthropic Economic Index, job_exposure.csv (data released under CC-BY)
- Anthropic Economic Index, Labor market impacts of AI: A new measure and early evidence (Massenkoff and McCrory), Mar 5, 2026
- Eloundou, Manning, Mishkin and Rock, GPTs are GPTs (2023)
- Felten, Raj and Seamans, AIOE data repository (no licence published; not used)
- Indeed Hiring Lab, AI tracker (CC BY 4.0)
- Indeed Hiring Lab, job postings tracker (CC BY 4.0)
- International Federation of Robotics, World Robotics 2026 press release
- O*NET 31.0 Database, USDOL/ETA (CC BY 4.0)
- O*NET licence terms
- OpenAI, GPTs-are-GPTs data repository (MIT licence)
- SiliconANGLE, Oct 21, 2025
- TechRepublic, Sep 9, 2025
- Windows Latest, Nov 11, 2026, citing Bloomberg and TechCrunch, Nov 10, 2026