# What AI Is Doing to White-Collar Jobs in Your Company Right Now | P-CATION Blog

> What AI is doing to office jobs: New research shows how white-collar work is changing.
> Source: https://p-cation.de/en/blog/what-ai-is-doing-to-white-collar-jobs/
> Language: en

[Back to blog](https://p-cation.de/en/blog/)

# What AI Is Doing to White-Collar Jobs in Your Company Right Now

What AI is doing to office jobs: New research shows how white-collar work is changing.

**Published:** April 28, 2026

**Updated:** April 29, 2026

**Author:** P-CATION Redaktion

News
#
AI in daily business operations
#
AI for companies
#
AI in SMEs
#
Optimize business processes
![Office workplace related to AI and modern knowledge work](https://p-cation.de/_astro/hero.ClojrUel_Z1olqoG.webp)

Anyone walking through an office today will rarely see the change at first glance. No job title disappears overnight. No team is suddenly replaced. And yet something is changing at the core of many white-collar jobs. First, someone has a meeting summarized. Then a research task becomes a prompt. Later, a system answers standard questions faster than a person could gather the information. The job is still there. But parts of it already feel different.

That is what makes Anthropic’s new research so interesting. It does not simply try to answer whether AI “replaces jobs”. It looks more closely: What could large language models theoretically already do, and what is actually being used in practice? This produces the familiar chart with the blue and red areas. Blue represents theoretical task coverage by LLMs. Red represents observed usage in real work contexts. The central message is sober: actual practice is still far below what is technically possible.

## What Anthropic Actually Measured

Anthropic calls the new metric observed exposure. It attempts to combine three things: the theoretical capabilities of language models, real usage data, and the question of whether AI is used more as support or more as automation. Automated and clearly work-related uses are weighted more heavily than purely supportive or unclear use cases. That matters because the metric does not merely estimate what AI might be able to do one day. It observes how work is already being touched today.

![Chart showing theoretical and observed AI coverage by occupational groups](https://p-cation.de/_astro/anthropic-observed-exposure.CiPwj6I1_ZTDt5R.webp)

The chart is powerful precisely because it shows two truths at once. On the one hand, theoretical reach is surprisingly broad. It is especially high in areas such as Computer & Math, Business & Finance, Management, Legal, and Office & Admin. On the other hand, actual usage remains much smaller in almost all categories. In other words: many tasks are already technically reachable, but they have not yet been systematically translated into everyday work.

## What the Study Says About White-Collar Work

For office jobs, this is not a side issue. Wherever work is strongly based on language, structure, analysis, documentation, and knowledge processing, the gap between capability and usage is especially relevant. Anthropic’s earlier Economic Index paper had already pointed in this direction: only about 4 percent of occupations showed AI use across at least 75 percent of their tasks, but about 36 percent of occupations already showed AI use across at least 25 percent of their tasks. Change therefore does not begin where an entire occupation “disappears”. It begins much earlier, at the level of individual task bundles.

The distinction between augmentation and automation is even more revealing. In the Economic Index paper, 57 percent of observed usage was supportive collaboration with AI and 43 percent was more automation-oriented usage. That sounds like a detail, but it is decisive for interpretation. The research does not say: AI makes people in office jobs unnecessary. It says something closer to this: in many cases, AI first changes the way people work before it fully takes over entire activities.

The distribution by qualification level is also interesting. Anthropic sees the highest usage not at the two extremes, meaning neither in very simple nor in the most specialized roles, but especially in Job Zone 4. These are roles with substantial preparation, often at bachelor’s degree level. According to the paper, usage there is 50 percent higher than one would expect based on the general labor market distribution. This zone covers many classic white-collar functions: well-trained knowledge work with clear language- and analysis-based tasks.

## Why This Is More About Tasks Than Jobs

This is probably the most important point in the whole topic. When people talk about AI, it often sounds as if entire occupations are competing against each other: does the job remain or disappear? The research shows a different picture. It is not the occupation as a whole that first comes under pressure. It is certain building blocks inside it.

This includes summarizing information, preparing answers, structuring knowledge, drafting content, comparing options, or condensing large amounts of text. These are tasks that happen constantly in office jobs, often between other responsibilities, and together they consume a surprising amount of time. This is exactly where AI connects particularly well.

Anthropic is careful in its wording. The researchers have so far found no systematic increase in unemployment in highly exposed occupations since late 2022. At the same time, they see signs that hiring of younger employees in more exposed occupations may have slowed. That is not proof of broad displacement, but it is a signal that the labor market may already be moving differently at the edges than many debates suggest.

## What Is Often Misunderstood

The wrong reading would now be: “Then not much is happening yet.” That would be too comfortable. The gap between theoretical capability and observed usage does not mean that the topic is still far away. It means that the technology is moving faster than organizations. The bottleneck today is often not the model, but processes, responsibilities, data access, rules, and trust. In short: many companies are not facing an AI capability limit, but an implementation limit inside their own organization.

For white-collar work, this has a clear consequence. The value of classic routine decreases. Anyone who mainly passes on information, polishes wording, or gathers knowledge comes under more pressure than someone who interprets, prioritizes, takes responsibility, and connects decisions with context. The work does not automatically become smaller. Often it becomes more demanding. Speed rises, comparability rises, and with it the expectation that standard work should no longer create as much friction as before. This conclusion is an interpretation of the study results, not a direct statement by the authors. But it follows logically from the observed shift toward tasks that can be structured and processed through language.

## A Short Look at Germany

For Germany, this picture fits surprisingly well. According to KfW, 20 percent of mid-sized companies now use AI; among larger mid-sized companies, the figure is 36 percent. At the same time, the ifo Institute shows that 64 percent of employees have used AI at least once, but only one in five uses it regularly at work. According to ifo, two thirds of the main applications were not introduced by the company, but used on employees’ own initiative. That points to a change that has already started, but is still not cleanly organized in many companies.

## What to Take Away from the Study

The calmest and probably most honest reading is this: the office job is not disappearing. It is being reorganized.

Not every role is affected equally. Not every task is immediately worth supporting with AI. And not everything that is theoretically possible will already make economic sense tomorrow. But the direction is clear. Wherever work consists strongly of writing, searching, summarizing, structuring, analyzing, and passing on information, AI is already changing the logic of daily work. Quietly at first, then noticeably.

Anyone who takes this seriously does not need to fall into alarmism. It is enough to look more closely. Not at occupations as fixed titles. But at tasks, friction, search time, knowledge access, and the question of what is still being done by hand in your own company simply because it grew that way historically.

That is where the real transformation begins.

Source: [Anthropic Research, Labor Market Impacts of AI: A New Measure and Early Evidence](https://www.anthropic.com/research/labor-market-impacts)

Are you wondering which tasks in your office areas could already run differently, faster, or automatically today? Then let us look at your workflows together and find out where meaningful pressure for change exists in your company and where real relief is possible. [Schedule an initial consultation](https://calendar.google.com/calendar/u/0/appointments/schedules/AcZssZ0319Na8C9fyYTJHz8fV8WABdTin4c9jMtlP_ROYgAlegmUR49vptziTkbuivZF-tDl527Fq2l6)