Anthropic Releases 2030 Scenario Suggesting Programming and Customer Service Jobs May Transition to Electricians and Nurses
Anthropic's economic team has launched a U.S. economic scenario browser for 2030, accompanied by technical reports from Korinek, Stanford scholar Jones, and others. The model breaks down occupations into O*NET tasks and presents three paths—mild, significant, and extreme—based on capabilities and adoption rates, without asserting that any industry will necessarily disappear by 2030.
In the significant and extreme scenarios, knowledge jobs face large-scale automation, with examples suggesting that coding and call center positions may transition to lower-exposure occupations like electricians and nurses. The report also notes the difficulties of job transition: willingness, skills, and re-employment delays could increase unemployment. In the extreme scenario, cognitive employment without AI is projected to decline by about 21.5%, with a knowledge job unemployment rate of about 17.9% and an overall unemployment rate of about 11.9%; labor's share of GDP is expected to drop from about 60% to about 45%; and non-knowledge job wages could rise by about 33.6% relative to a no-AI path. In the significant scenario, economic growth is expected to roughly double, with knowledge job wages remaining relatively stable. The mild scenario is close to the impact of the internet, with wages remaining stable.
Earlier economic indices ranked computer programmers and customer service representatives among the highest exposure occupations, at approximately 75% and 70%, respectively. CEO Dario Amodei previously stated that entry-level white-collar jobs could disappear by half by 2030, with unemployment rates potentially reaching 10% to 20%. External commentary suggests these figures are more aligned with the model's extreme path. A median expectation from a survey of over 10,000 respondents is close to the significant scenario, with about 10% of respondents anticipating extreme outcomes. The compensation gap for knowledge job income under the extreme path is estimated to be about 9% of GDP.
Mechanically, wages for tasks being automated shift towards capital, primarily computational power; as design and licensing accelerate, demand for on-site construction and nursing increases. Buyers are institutions that need to show policymakers and investors the impact range, while sellers are task models calibrated by Anthropic using their own product usage data. The narrative around funding shifts from "universal wage increases" to stagnation or decline in knowledge jobs, with a scarcity premium for physical and nursing jobs. Beneficiaries include electricians, nurses, and construction workers—occupations less likely to be replaced by models—while coding and telephone customer service jobs, which produce text, face pressure, along with finances that require large-scale transfer payments to offset losses.
Source: Public Information
ABAB AI Insight
Anthropic is simultaneously selling Claude Code to accelerate coding automation while releasing a macro model that includes programmers on the transition list, translating product penetration into labor market equations. Amodei previously stated at Davos that the model could complete most or even all tasks of software engineers within 6 to 12 months, yet the company's engineering workforce remains predominantly high-paid technical staff. The scenario browser marks the CEO's most severe unemployment rate range as the extreme path, effectively adding confidence labels to external statements using internal models rather than issuing job training manuals.
The capital path indicates that after tasks are automated, compensation shifts from wage accounts to the party owning the clusters. The expansion of data centers simultaneously raises the prices of electricians, welders, and plumbers, creating a scenario where "the model replaces those who write code and then hires those who connect wires." Nursing jobs are included in the examples because bedside labor and legal responsibilities have yet to enter the realm of scalable deployment by robots. The delay in job transition is framed as an unemployment mechanism: it is not that the total number of jobs drops to zero immediately, but rather that the months spent for programmers to obtain electrician licenses become statistically counted as unemployment.
Similar narratives were seen with the early internet's impact on travel agencies, ATMs on tellers, and the current competition for blue-collar wages driven by data centers. The industry phase indicates that knowledge work is moving from enhancement to an increasing share of replaceability, while physical and nursing jobs are being reclassified from "low skill" to non-outsourcable scarcity. The model company provides three curves to prepare adjustable parameters for regulators and Congress, rather than issuing individual job transition notices.
Structurally, this reflects a reconstruction of income distribution under technological substitution: labor's share declines, capital's share rises, and on-site work becomes more expensive due to AI accelerating upstream design. The mechanism is that the marginal costs of text and code tasks approach reasoning prices, while the marginal costs of physical and compliance tasks remain tied to apprenticeship durations and licenses; job transition friction temporarily manifests as knowledge job unemployment rates, rather than everyone becoming wealthy simultaneously.
ABAB News · Cognitive Law
The model first displaces text jobs, then raises wages for on-site jobs.
The delay in job transition is the production function of the unemployment rate.
The fastest-growing scenarios often see the most significant drops in labor share.