Most occupations exposed to AI are not disappearing. They are being taken apart at the task level and reassembled, with some tasks handed fully to automation, others turned into human-AI collaboration, and a smaller set of tasks becoming more valuable precisely because AI cannot do them. Anthropic’s Economic Index puts collaborative, augmented use of AI ahead of full automation in the majority of tracked work, even as the automation share continues to grow. This is the pattern every function leader planning workforce capacity for 2027 needs to plan against: role-by-role redesign, not blanket elimination, with the sharpest real cost falling on hiring for the youngest workers rather than on cutting the workers already in seat.
Exhibit 01
52%
Share of Claude conversations classified as augmentation, not automation, Nov 2025
-19%
Employment shortfall for workers aged 22-25 in highly AI-exposed occupations, vs. expected trend
1.3M
New AI-enabled jobs created globally over two years, LinkedIn platform data
2.2x
Pay premium for AI-related job postings vs. non-AI roles, US, 2026
Source: Anthropic Economic Index, Jun 26, 2026. Stanford Digital Economy Lab, Aug 11, 2026. LinkedIn, Jan 13, 2026 and Aug 17, 2026.
Chapter 01
What happened
The dominant public narrative treats AI’s effect on work as a single dial running from “jobs safe” to “jobs gone.” The data does not support a single dial. Anthropic’s Economic Index, built from roughly one million classified Claude conversations, found automation briefly overtook augmentation in an August 2025 sample (49% automated to 47% augmented) before reversing sharply back to an augmentation-led split by November 2025, when 52% of conversations were classified as augmented against 45% automated. The balance is not sliding steadily toward automation; it is oscillating, and collaboration, not replacement, remains the more common mode of AI use as of the most recent data.
A second, independent measurement points the same direction. Microsoft’s 2026 Work Trend Index, drawing on more than 100,000 analyzed Copilot chats plus a 20,000-worker, 10-country survey, found 49% of conversations supported cognitive work, meaning analysis, problem-solving, and thinking that once required deep expertise, with the remainder split across working with people, producing outputs, and finding information. Microsoft separately found 58% of AI users report producing work they could not have a year ago, a figure that rises to 80% among its most advanced “Frontier Professionals.” Roughly even odds between cognitive and other modes of AI use is a redesign signal, not an elimination signal: if AI were simply removing roles wholesale, usage would skew overwhelmingly toward the most mechanically transactional tasks, since those are the easiest to hand off entirely. It does not.
Worth noting
The obvious skeptical read: task-redesign language is corporate cover for ordinary layoffs. There is real evidence for a version of this. Axios reporting, published August 20, 2026, found executives are now deliberately distinguishing “AI is reshaping how we work” from “AI is replacing workers” in public statements, a documented shift in communications strategy. A single LinkedIn commentator’s observation, not a verified dataset, claims most 2026 layoff memos cite AI regardless of underlying financial condition, citing Intuit and Coinbase. And Reuters’ investigation into internal Meta documents found a real gap between an internal restructuring plan, “Project OT,” which explored cutting team sizes by up to 60% before Meta canceled the planned second wave, citing internally observed shortfalls in the productivity gains the plan assumed. Messaging games are real. But the report’s load-bearing finding, that there is no evidence of widespread economy-wide displacement, comes from Stanford’s ADP administrative payroll data covering millions of actual paychecks through June 2026, not from press releases or survey self-report. A company can spin a layoff; it cannot spin the aggregate national payroll record. The redesign pattern holds independent of any single company’s messaging choices.
Chapter 02
Where the redesign is landing
Stanford’s Digital Economy Lab, using ADP payroll data covering millions of US workers, states plainly that it finds no evidence of widespread, economy-wide job displacement from AI. That calm aggregate picture, however, sits on top of a sharp and widening divergence by age. Employment for workers 22 to 25 in the most AI-exposed occupations is 19% below where it would sit had it tracked less-exposed peers, up from a 15% gap recorded a year earlier. The mechanism is reduced hiring, not increased firing, a distinction the lab’s director, Erik Brynjolfsson, has restated directly: the gap is concentrated in occupations like software development and customer service, and 34 straight months of year-over-year contraction in youth AI-exposed employment have now been recorded.
This single-source-family caveat matters and belongs on the page rather than in a footnote: the 19% figure and its year-over-year growth are documented twice, once in the Stanford Digital Economy Lab’s own published report and once in Brynjolfsson’s restatement of the same underlying ADP research program. That is a strengthened restatement of one primary source, not two independent datasets, and no separate third-party study, from BLS, Challenger, or an academic group outside Stanford’s ADP partnership, was found replicating the exact figure. The direction of the finding is well corroborated by adjacent evidence; the precise number should be read as coming from one research program, stated once with more detail the second time.
Redesign, not elimination, is visible directly in task-level data too. Anthropic’s own task-mapping tool classifies specific occupations against O*NET task categories to show which tasks within a role skew toward automated use and which skew toward augmented, collaborative use, a split that varies task by task within the same job title rather than applying to an occupation wholesale. Stanford’s SALT Lab, auditing 1,500 workers across 104 occupations, operationalizes the same underlying idea with a four-zone framework: an “Automation Green Light” zone where workers want AI to take over and it is currently capable of doing so, an “Automation Red Light” zone where AI is capable but workers do not want the handoff, an “R&D Opportunity” zone where workers want AI help that current tools cannot yet deliver, and a residual “Low Priority” zone. None of the four zones is “job eliminated.” All four describe a task, not a job title, moving to a different place in a role’s overall composition.
| Task category |
What the evidence shows |
| Automated |
Structured, transactional tasks such as data entry and document handling skew heavily toward full automation in Anthropic’s O*NET-based task-mapping data, and toward BLS’s projected declines for procurement clerks and customer service representatives through 2034. |
| AI-augmented |
Analysis, evaluation, and creative-thinking tasks skew toward collaborative, augmented use rather than automation, consistent with Microsoft’s finding that roughly half of Copilot use supports this kind of cognitive work. |
| Newly human-critical |
Stanford SALT Lab’s “R&D Opportunity” zone, tasks workers want AI help with but current tools cannot yet deliver, marks where judgment and oversight stay human by necessity, not preference alone. |
| Newly created |
LinkedIn’s Jobs on the Rise 2026 ranks AI Engineer as the fastest-growing US role, with VP of AI postings up roughly sixfold and Forward Deployed Engineer postings, a role that barely existed three years ago, growing several hundred percent year over year. |
This framework synthesizes task-level classifications across four independent research programs, Anthropic’s Economic Index, Microsoft’s Work Trend Index, Stanford SALT Lab, and LinkedIn’s Economic Graph. It is an analytical grouping built for this report, not a single named index.
The scale of the “newly created” category shows up clearly in BLS’s Employment Projections 2024-2034, published July 16, 2026: data scientists are projected to grow 33.5% (82,500 new jobs), information security analysts 28.5% (52,100 new jobs), and software developers 15.8% (267,700 new jobs), all well above the average projected growth for all occupations. Adjacent, administrative-facing roles are not disappearing under this same projection, they are simply growing more slowly: medical secretaries and administrative assistants are projected to grow 4.2% (35,300 new jobs) over the same decade. That is a redistribution of growth rate, not a reversal into decline, and it directly complicates a framing where AI-adjacent administrative work is treated as doomed.
The flip side
The same dataset that documents the youth-hiring gap also documents where demand is genuinely strong: AI-related job postings have roughly doubled in the US since 2023, and the typical AI-related posting carries about $177,000 in compensation versus $80,000 for a comparable non-AI role, a 2.2x premium, per LinkedIn’s Economic Graph research. Globally, 1.3 million new AI-enabled jobs have appeared over two years, including more than 600,000 new AI-enabled data center jobs in the most recent year alone. The youth-hiring gap is real and specific to entry pathways in a narrow set of exposed occupations; it is not evidence that AI-adjacent work broadly pays less or hires less overall.
Case Study 01
C3.ai · Enterprise software · United States
C3.ai: a distressed turnaround, not a clean redesign case
A real, quoted, named example of an AI company restructuring around its own AI products under financial pressure, useful as a boundary case for where this report’s core thesis does and does not apply.
The context
Tom Siebel returned as Chairman and CEO of C3.ai in early 2026 amid what he called “staggeringly disappointing” company performance, with an explicit mandate to turn the business around. Sales had, in his words, “fallen off a cliff” over the preceding five quarters.
What changed
Within his first months back, Siebel expanded and accelerated a restructuring already begun by his predecessor, reorganizing sales, product, and services under new leadership and reducing headcount by approximately 35% across all organizations, from roughly 1,075 to about 700 people.
The evidence
On C3.ai’s earnings call for the fourth quarter and full fiscal year 2026, for the quarter ended April 30, 2026, held June 3, 2026, Siebel told analysts: “We have reduced headcount by approximately 35% across all organizations… We have taken almost $135 million annual operating cost for the business structure.” Describing the scope of the reorganization, he said the company had “restructured the company, have restructured sales… restructured products… restructured services.” Source: C3.ai fiscal Q4 2026 earnings call transcript, The Motley Fool, accessed September 10, 2026.
What transfers
- A leadership change and a financial turnaround can drive a large headcount reduction on their own, independent of any task-level automation calculation; the two forces are easy to conflate in public reporting.
- Cost-discipline narratives and AI-strategy narratives are frequently bundled together in public company communication; workforce planners evaluating a peer’s “AI-driven” cuts should separate the two before drawing conclusions for their own organization.
- A large headcount reduction paired with a pivot toward an AI-native product strategy is a real, occurring pattern in 2026, and should be treated as a distinct workforce-risk category from the gradual, task-level redesign documented in the rest of this report.
Limits
This is a distressed-turnaround case, not an isolated instance of AI-driven task redesign. The 35% headcount reduction is entangled with a company-wide restructuring under a newly returned CEO, including sales-execution problems the company itself attributed to management discipline rather than product quality or market size. C3.ai has not published a task-level automation-versus-augmentation breakdown for its retained workforce, so this case cannot be read as direct proof of this report’s central “tasks redesigned, not jobs eliminated” thesis at the individual-worker level. It should be read as a real, named, quoted example of an AI company reorganizing under cost pressure, current as of the quarter ended April 30, 2026.
Career takeaway:
A company’s own “AI transformation” framing for a layoff is not, by itself, evidence of task-level automation; the underlying financial and strategic context needs separate scrutiny.
Sources: C3.ai fiscal Q4 2026 earnings call transcript, quarter and fiscal year ended April 30, 2026, held June 3, 2026, accessed September 10, 2026.
Watch out
The most common misreading of this data is collapsing “no widespread economy-wide displacement” into “no real risk anywhere.” The two findings coexist: the aggregate national payroll picture is calm while a specific, well-documented, and widening gap exists for workers aged 22 to 25 in a narrow set of highly exposed occupations. A workforce plan built only on the aggregate number will miss the entry-level pipeline problem entirely, since that risk shows up as reduced hiring in specific functions, not as a visible layoff line item anywhere in the data.
Frequently asked questions
Is AI eliminating jobs or just changing them?
+
Primarily changing them. Stanford’s Digital Economy Lab, using ADP payroll data through June 2026, finds no evidence of widespread, economy-wide job displacement, while task-level data from Anthropic and Microsoft shows collaboration remaining the dominant mode of AI use.
Which workers are actually losing ground because of AI?
+
Workers aged 22 to 25 in highly AI-exposed occupations, per Stanford Digital Economy Lab. Their employment sits 19% below expected trend, up from 15% a year earlier, driven by reduced hiring, not increased layoffs, concentrated in software development and customer service.
What kinds of tasks are being fully automated right now?
+
Transactional, structured tasks like data entry, document handling, and routine records work. Anthropic’s O*NET-based task-mapping tool shows these tasks skewed heavily toward automated use, while analysis, evaluation, and creative-thinking tasks in the same roles skew toward collaborative, augmented use instead, per the same dataset.
What new roles has AI actually created?
+
AI Engineer ranks first on LinkedIn’s Jobs on the Rise 2026 list. VP of AI postings are up roughly sixfold, and Forward Deployed Engineer, a role helping organizations implement AI systems, is now the third most common AI occupation in US postings.
What should workforce planning teams account for given this task-level redesign?
+
Plan for the entry-level hiring pipeline separately from the incumbent workforce. Aggregate headcount stability can mask a real, growing gap in junior hiring for AI-exposed functions like software development and customer service, which will not show up in a standard attrition or layoff dashboard.
Sources
Anthropic Economic Index, “Cadences” report and January 2026 report (Nov 2025 data), published Jun 26, 2026 and Jan 2026. Anthropic Economic Index task-mapping tool, anthropic.com/economic-index, accessed Sep 2026. Microsoft 2026 Work Trend Index, published May 5, 2026. Stanford Digital Economy Lab, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence,” published Aug 12, 2026 (data through Jun 2026). Erik Brynjolfsson (Stanford Digital Economy Lab), LinkedIn post citing ADP Research/Stanford data, published Aug 11, 2026. Stanford SALT Lab, “Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the U.S. Workforce,” arXiv, 2025-2026. BLS, “Artificial intelligence, information technology, and employment, 2024-34,” published Jul 16, 2026. LinkedIn Economic Graph research on AI job postings and compensation, reported Aug 2026. LinkedIn, “Building a Future of Work That Works,” Davos press release, published Jan 2026. LinkedIn, “Jobs on the Rise 2026,” published Jan 6, 2026. World Economic Forum, “Artificial Intelligence and the Future of Entry-Level Work,” 2026. Axios, “CEOs shift messaging around AI and layoffs,” published Aug 20, 2026. Reuters investigation, “Mark Zuckerberg had a bold plan to replace Meta staff with AI,” published Aug 26, 2026. C3.ai fiscal Q4 2026 earnings call transcript, quarter and fiscal year ended Apr 30, 2026, held Jun 3, 2026, accessed Sep 10, 2026. Brit Morin, LinkedIn post, published Jun 14, 2026, cited as an anecdotal, non-dataset observation.
Free Tool · AI Job Risk Calculator
Which of your role’s tasks fall into which category?
See where a specific role sits against the automated, augmented, human-critical, and newly created framework above.
Check free ->