AI’s Impact on the Workforce: Tasks, Roles, and What Hiring Data Shows

AI’s impact on the workforce is easiest to understand at the task level, according to Anthropic’s research on employment, which examined more than 11,000 activities and mapped their exposure across more than 900 O*NET occupations. The study found that writing, coding, research and information processing face greater AI exposure than physical work, while hiring data shows slower growth in some highly exposed categories and more demand for AI-related skills.

What did the study measure?

Anthropic’s research combines task-level analysis with occupation exposure scores and labor-market data. That approach gives a more precise picture of AI exposure than looking only at job titles, since two workers with similar titles may spend very different amounts of time on automatable tasks.

• Task-level analysis: More than 11,000 individual work activities were scored for the share AI could perform or assist with, using expert annotation and model evaluation.
• Occupation exposure: The task scores were mapped to more than 900 O*NET occupations, producing an exposure score for each role.
• Labor-market signals: The researchers compared job postings, required skills and role categories before and after widely available AI assistants appeared.

A job title can suggest exposure to automation even when much of the position depends on judgment, relationships or physical work. A less technical role may still require substantial drafting, research or data-processing tasks.

Which tasks have the greatest exposure?

Cognitive and desk-based activities have the highest exposure to AI. These tasks often involve producing, organizing, or interpreting written information, and that gives AI systems material to work with.

• Reading, writing and information synthesis: Gathering text, summarizing sources, drafting responses and comparing documents score highly for automation and augmentation.
• Research and analysis support: Building briefs, compiling data and preparing first drafts of analysis are highly exposed, while human judgment remains central to deciding what is reliable or useful.
• Customer communications: Routine correspondence and ticket-style responses map to higher exposure because these tasks often follow repeatable patterns.
• Basic coding: Well-defined code generation is highly exposed, although broader software work can also require architecture, debugging and judgment about system requirements.

Tasks that depend on physical presence, motor skill or in-person care rank lower. That does not mean those jobs are changing more slowly. It means AI exposure, as defined by this study, is lower because the systems analyzed are digital and cannot perform the full physical task.

Which roles score highest?

Occupations dominated by writing, research, coding and information processing sit toward the top of the exposure distribution. Anthropic’s analysis identifies these broad groups as especially exposed:

• Office and administrative support roles involving scheduling, document preparation and correspondence.
• Business and financial operations roles centered on text, models and research synthesis.
• Computer and mathematical roles focused on code and structured data work.
• Sales roles where much of the time goes to prospecting, drafting and customer-management-system updates rather than closing transactions.

These are occupation-level patterns, not predictions that every job within those categories will be automated. A role’s exposure depends on the activities that make up its daily work. The report also places roles built around physical execution, care delivery, repair and skilled trades at lower exposure because human presence and physical ability remain necessary.

What has changed in job postings?

Anthropic’s labor-market analysis compares the period after widely available AI assistants with the earlier baseline. The report finds slower growth in job postings within some highly exposed categories than in lower-exposure categories. Some highly exposed roles show declining growth, but the pattern is a reduction in the rate of expansion rather than a reported collapse in total employment.

Skill requirements also show a shift toward AI literacy. Postings more often mention prompt engineering, model evaluation and the ability to use AI tools within a workflow. The study identifies new role categories such as AI-augmented analyst, prompt engineer and AI operations specialist.

They do not prove AI caused each change in an occupation. Other economic conditions can affect posting growth, and the report tracks how the data shifts after AI tools become widely available.

What does broader research say about employment?

Other studies add context to the workforce findings. A Budget Lab analysis published October 1, 2025 examined the first 33 months after ChatGPT’s November 2022 release. The authors found that the occupational mix was changing faster than in earlier periods, though not by a large amount. The mix change also predated widespread AI introduction, and the researchers found no sign that exposure, automation or augmentation measures were related to changes in employment or unemployment during that period.

The same analysis cautions that 33 months is a short period for measuring a technology that may reshape work over decades. Its authors say the results are a snapshot, not a prediction, and plan to monitor the trends over time.

MIT Sloan reported on research using about 58 million LinkedIn profiles and 14 million job postings, combined with roughly 20,000 O*NET work activities. The researchers found that when AI could perform most tasks in a role, that role’s share of employment within a company fell by about 14%. When AI affected only a few tasks, employment in the role could grow, partly because workers could spend more time on activities outside AI’s reach.

The study, published by MIT Sloan on October 9, 2025, also found that AI-exposed roles were concentrated in higher-paying information-processing and analysis work. From 2010 through 2023, highly exposed roles did not experience job losses relative to other roles. The researchers associated extensive AI use with about 6% higher employment growth and 9.5% more sales growth over five years, although their findings predate the widespread rise of generative AI.

A Stanford policy brief similarly reports that aggregate AI-related employment effects are likely small at present. Its analysis, using exposure measures and labor-market data through 2026, says unemployment among workers in the most exposed occupations rose by 0.77 percentage points since 2022, compared with a 0.85-point rise among the least exposed workers.

The brief also reports that employment among firms adopting enterprise AI grew by 10% in the two years after adoption, driven mainly by firms with the highest AI spending per employee. It says evidence points to role consolidation and hiring avoidance in some occupations, while also noting that a difficult market for recent graduates may partly reflect AI.

What can workers and employers do with this information?

The task-level approach offers a practical way to read a job description. Workers can list the activities they perform, identify the tasks that AI could perform or assist with, and decide which parts of the role need stronger human judgment, relationship skills or review.

That review work has practical value because an AI-generated draft can be wrong, incomplete or based on weak sources. The research findings do not prove that judgment will protect every job from change. They do show why the non-automated part of a role, such as checking an answer, negotiating with a client or making a decision under uncertainty, may remain important as more of the routine work is supported by AI.

Employers can use the same method when redesigning a position. Instead of beginning with a title, they can compare the task list, identify work that can be supported by AI, and preserve clear responsibility for review, customer communication and domain decisions. The research also gives a reason to track skill language in current postings, since AI-related requirements are appearing in hiring descriptions alongside the established skills for each occupation.

FAQ

What did the Anthropic study find about AI and jobs?

Anthropic’s study found that AI exposure is concentrated in writing, research, coding and information-processing tasks. Office and administrative support, business and financial operations, and computer and mathematical roles were among the more exposed occupation groups. The report also found slower growth in some highly exposed job categories and increased emphasis on AI-related skills.

Which jobs are most exposed to AI?

Roles dominated by text, research, structured analysis and code score highest in the study. This includes many administrative, analyst and software positions. Roles built around physical execution, in-person service, repair and skilled trades score lower because physical presence and motor skill remain necessary.

Is AI causing job losses or changing roles?

The research shows changes in tasks, skills and job descriptions, including slower growth in some highly exposed postings rather than a sharp drop in total employment. Broader research from Yale, MIT Sloan and Stanford describes the current evidence as limited and mixed, with possible effects on particular groups and occupations that may be difficult to see in economy-wide data.

SEOScanPro

SEOScanPro runs a full technical audit of a site and shows the measured result behind every check. Open the site audit tool.

See our editorial disclaimer for how our articles are produced.