For the past two years, the easiest way to talk about AI and jobs was to talk about programmers. That made sense. Software workers were the first large professional group to feel generative AI changing daily work in visible ways, from code generation to debugging to agent-assisted workflows.
But the labor story is no longer narrow enough to stay there. Anthropic's March 24, 2026 Economic Index report found Claude.ai usage becoming less concentrated and management-related work rising from 3% to 5% of traffic, driven by tasks such as preparing investment memos and answering customer questions.[^2] That does not prove a full labor-market shift on its own, but it is a concrete sign that AI use is broadening beyond classic software workflows.
Anthropic's research still shows that coding remains one of the clearest early zones of AI adoption. In an April 2025 analysis of 500,000 coding-related interactions, the company found that 79% of Claude Code conversations were categorized as automation rather than augmentation, compared with 49% on Claude.ai.[^1] The same analysis found heavy use around user-facing application work and suggested that simpler app and interface tasks could face disruption sooner than deeper backend work.[^1] But Anthropic also warned that software may be a leading indicator rather than the whole story.[^1]
That matters for non-tech workers because the broader question is no longer just whether coders automate faster. It is whether administrative, customer-facing, education, and management tasks start getting reorganized in the same direction.
In its March 24, 2026 Economic Index report, Anthropic said Claude.ai use had diversified beyond earlier concentrations.[^2] It also said the cumulative estimate from its prior reporting that about 49% of jobs had seen at least a quarter of their tasks performed using Claude had "barely changed" in the new data pull.[^2] Separately, the report found that the average value of tasks done on Claude.ai had decreased slightly, largely because personal queries rose and coding shifted toward the API.[^2]
That reported diversification does not by itself prove that AI has fully spread into a broad layer of lower-wage work. The inference is narrower and more defensible: if use cases are widening, management-related tasks are increasing, and customer-service workflows are showing higher automation exposure in Anthropic's API data, then the AI labor story is moving beyond a software-only frame.[^2]
That shift matters because it changes how the AI labor story should be framed. This is not only a story about whether software engineers lose jobs. It is increasingly a story about how work gets reorganized across a much larger share of the economy.
The bigger issue is job redesign, not only job loss
Public debate still defaults to a blunt question: will AI replace workers?
That question is too crude to explain what the data is actually showing. Anthropic's work points to a mix of augmentation and automation rather than a clean one-direction story.[^1][^2] Even in coding, where AI use is unusually intense, the pattern is not simply that humans disappear. It is that some tasks become more automated, some become more collaborative, and the structure of the job changes around them.
That framing becomes even more important once AI moves beyond tech roles. Many occupations will not vanish in one dramatic wave. Instead, the content of jobs will shift. Some entry-level responsibilities may be absorbed by AI tools. Some workers will be expected to supervise, edit, verify, or orchestrate machine output. Others will be pushed toward more interpersonal or situational work that is harder to standardize.
That is still disruption. In some ways, it is harder to navigate than outright replacement because it creates uncertainty without a clean endpoint. Workers may keep the title while losing the old path for building expertise, moving up, or proving value.
Why the pressure is spreading beyond coders
The most useful counterweight to a tech-only framing comes from ETS's 2026 Human Progress Report, released April 1, 2026. Drawing on responses from more than 32,000 adults across 18 countries, ETS describes a workforce that is adapting, but without much clarity about where adaptation is supposed to lead.[^3]
The findings are striking. ETS said 77% of workers believe job security now requires continuous evolution, and the same share said they are proactively building new skills.[^3] But 71% said they cannot envision the future jobs those skills are preparing them for.[^3] Sixty percent said they feel pressured to adopt AI tools before they are ready, and 73% said it is difficult to know what level of AI literacy employers expect.[^3]
That is a broader social signal than a coding-productivity study. It suggests the AI labor shock has moved from a specialist workflow story into a mass uncertainty story.
This is also where the non-elite dimension becomes harder to ignore. ETS said disparities persist for women, older workers, rural populations, and people without credentials.[^3] It also found that 85% of workers see credentials as essential for career survival, while only 45% say they have access to credentialing programs.[^3] That gap matters because AI disruption does not hit workers evenly. People with clearer access to training, better signals from employers, and stronger institutional support are much more likely to adapt successfully than workers asked to reinvent themselves on their own.
So the emerging divide is not just technical versus nontechnical. It is increasingly structured by who can convert adaptation pressure into credible opportunity.
Early adopters may benefit first, which can widen inequality
Anthropic's March 2026 report adds another layer to that risk. The company found that more experienced Claude users tend to bring more complex, more work-related tasks to the system and are more likely to get successful responses.[^2] Anthropic explicitly noted that these patterns could deepen labor-market inequalities if effective AI use depends on complementary skills that some workers acquire earlier or more easily than others.[^2]
That is an important point because it shifts the question from "Who is exposed?" to "Who is able to benefit?"
Workers in technical and knowledge-heavy roles may be the first to face disruption, but they may also be among the first to learn how to use AI productively enough to capture more upside. Workers outside those environments may face the pressure to adapt without the same institutional support, experimentation time, or training infrastructure. That is one reason the labor impact should not be understood as a simple race between humans and machines. It is also a race between workers with strong adaptation scaffolding and workers without it.
What workers and employers should do next
Workers should treat AI literacy as job insurance, but in a targeted way. The useful response is not to chase every new tool. It is to identify which parts of a role are easiest to automate, then build stronger skills in review, judgment, client interaction, domain expertise, and workflow design, the areas where human value is harder to commoditize.
Employers should make adaptation concrete instead of rhetorical. That means spelling out what AI literacy actually means for each role, protecting enough entry-level work for people to build expertise, and funding credentials or training paths that do not leave lower-support workers behind. If companies push AI adoption without redesigning career ladders, they risk creating a more efficient system that is also harder to enter and less fair.
What policymakers and employers are still underestimating
One risk in current AI debate is that organizations still talk as if adoption itself is the hard part. The harder part may be making adaptation legible and fair.
If employers increasingly expect AI literacy but cannot define what that means, workers are left guessing.[^3] If AI tools absorb chunks of entry-level work, companies may quietly weaken the ladder that used to train future mid-career talent. If access to credentials and reskilling remains uneven, AI could reinforce the exact inequalities that workforce policy is supposed to soften.[^3]
The AI labor shock is therefore spreading in two directions at once. It is spreading outward across occupations beyond tech. And it is spreading downward into the institutional layers that shape mobility: training, credentials, hiring signals, and early-career pathways.
That is why "job redesign" is the more useful phrase. It captures the fact that the disruption is broadening before the full employment effects are even visible. The real challenge is not only preventing displacement. It is making sure the new structure of work does not become more opaque, more unequal, and harder to enter.
Software workers were simply first to feel the change. They are no longer the only ones.
Sources
[^1]: Anthropic, "Anthropic Economic Index: AI's impact on software development," April 28, 2025, https://www.anthropic.com/news/impact-software-development
[^2]: Anthropic, "Anthropic Economic Index report: Learning curves," March 24, 2026, https://www.anthropic.com/research/economic-index-march-2026-report
[^3]: ETS, "Adaptability Revealed as the New Foundation of Job Security in the AI Age, According to 2026 ETS Human Progress Report," April 1, 2026, https://www.ets.org/newsroom/adaptability-revealed-as-new-foundation-of-job-security-in-ai-age-human-progress-report-finds.html
Published by themimic.io — tracking the humanoid robotics industry without the hype.