{"id":12182,"date":"2026-04-02T11:48:07","date_gmt":"2026-04-02T09:48:07","guid":{"rendered":"https:\/\/job.manualjobsearch.com\/?p=12182"},"modified":"2026-04-02T11:55:58","modified_gmt":"2026-04-02T09:55:58","slug":"ki-und-der-arbeitsmarkt-2025-wer-wirklich-gefaehrdet-ist-und-wer-nicht","status":"publish","type":"post","link":"https:\/\/job.manualjobsearch.com\/en\/ki-und-der-arbeitsmarkt-2025-wer-wirklich-gefaehrdet-ist-und-wer-nicht\/","title":{"rendered":"AI and the labour market 2025: Who is really at risk \u2013 and who isn't"},"content":{"rendered":"
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t
\nA recent Anthropic study provides real-time data for the first time, instead of forecasts. The results contradict everything previously believed to be known.<\/i>\n\n

No more forecasts \u2013 real data<\/h2>\nFor years, economists, management consultants, and technologists have predicted which professions will disappear due to artificial intelligence. The predictions were often dramatic \u2013 and just as often wrong. The panic surrounding outsourcing in the early 2000s cost hardly as many jobs as predicted. The wave of automation in industry proceeded more slowly and selectively than expected.

\n\nThe fundamental problem with these predictions has always been the same: they were based on what machines could theoretically achieve \u2013 not on what they actually do in everyday life. A language model can theoretically draft legal documents, create financial reports, and debug code. But how often does this actually happen? To what extent? In which professions?

\n\nA study by researchers at Anthropic, Maxim Masenkov and Peter Macroory, now answers precisely this question. They have developed a new metric \u2013 the so-called Observed Task Coverage (OTC). And the results are as astonishing as they are revealing.\n\n

The Method: Theory Meets Reality<\/h2>\nThe researchers combined three data sources:\n
    \n \t
  • O*NET \u2013 the US federal occupational classifier. This database breaks down every occupation into precise micro-tasks with time allocations.<\/li>\n \t
  • Theoretical AI performance evaluation \u2013 assessments of how much AI can accelerate tasks.<\/li>\n \t
  • Real usage data of the Claude model \u2013 which tasks people actually delegate to AI.<\/li>\n<\/ul>\n\nThe crucial difference: Instead of assumptions, actual behaviour was analysed.\n\n

    The Food Processor Paradox<\/h2>\nTo illustrate the core problem of earlier studies, the researchers use a vivid analogy:\n\n
    A food processor can theoretically cook a five-course meal \u2013 in practice, it's usually only used for chopping onions.<\/div>\n\nThe exact same applies to AI. The central question is not what it can do \u2013 but what is actually being used. The answer: we are still at the \"onion peeling\" stage.\n\n

    The figures: The gap between potential and reality<\/h2>\nAn example from practice:\n\n
      \n \t
    • Theoretical AI potential (IT professions): 94 % of tasks<\/li>\n \t
    • Actual use: only about 3 %<\/li>\n<\/ul>\n\nThis enormous gap arises from three central barriers:\n
        \n \t
      • Legal liability (e.g., medical, legal)<\/li>\n \t
      • Outdated IT infrastructure in companies<\/li>\n \t
      • Human control and decision-making processes<\/li>\n<\/ul>\n\n
        The limitation is not technological \u2013 but institutional.<\/div>\n\n

        Who is really at risk? The ranking<\/h2>\n\n
        Rang<\/th>Field of work<\/th>Observation cover<\/th>Main task<\/th><\/tr>\n
        1<\/td>Programmer<\/td>74,5 %<\/td>Code schreiben und warten<\/td><\/tr>\n
        2<\/td>Customer support<\/td>70 %<\/td>Communication<\/td><\/tr>\n
        3<\/td>Data input<\/td>67 %<\/td>Data processing<\/td><\/tr>\n
        4<\/td>Medical Documentation<\/td>~55 %<\/td>Create reports<\/td><\/tr>\n
        5<\/td>Marketing analysis<\/td>~50 %<\/td>Reporting<\/td><\/tr>\n
        6<\/td>Sales<\/td>~45 %<\/td>Offers & Follow-ups<\/td><\/tr>\n
        7<\/td>Financial analysis<\/td>~44 %<\/td>Forecasts<\/td><\/tr>\n
        8<\/td>Software Testing<\/td>~40 %<\/td>Test cases<\/td><\/tr>\n
        9<\/td>IT Security<\/td>~38 %<\/td>Threat analysis<\/td><\/tr>\n<\/table>\n\nAt the lower end: craft and physical occupations \u2013 often with 0 % AI coverage.\n\n

        The Paradox of Education<\/h2>\nA surprising result:\n\n
          \n \t
        • 17 % from the at-risk group have a master's degree<\/li>\n \t
        • Only 4.5 % of this group are barely affected by AI<\/li>\n<\/ul>\n\n
          The higher the qualification, the higher the AI exposure, often.<\/div>\n\nTasks that are purely cognitive \u2013 writing, analysing, structuring \u2013 are particularly well-suited for automation.\n\n

          Where are the mass layoffs?<\/h2>\nDie Antwort: Sie passieren indirekt.\n\nUnternehmen entlassen keine erfahrenen Mitarbeiter. Stattdessen: The answer: They happen indirectly.\n\nCompanies do not lay off experienced employees. Instead:\n
            \n \t
          • Senior staff become more productive<\/li>\n \t
          • Fewer junior positions are being created<\/li>\n<\/ul>\n\nData shows:\n
              \n \t
            • New hires (aged 22\u201325) in AI roles: \u221214% % since 2022<\/li>\n<\/ul>\n\n
              The door to the job market is quietly closing \u2013 not visibly.<\/div>\n\n

              Long-term effects<\/h2>\nStatistical models show:\n\n
                \n \t
              • +10 % AI coverage \u2192 \u22120.6 % employment growth<\/li>\n<\/ul>\n\nOver years, this adds up to significant effects \u2013 particularly in highly automated professions.\n\n

                Strategies: What to do?<\/h2>\n\n

                For experienced professionals<\/h3>\n
                  \n \t
                • Using AI as a productivity booster<\/li>\n \t
                • Automate workflows<\/li>\n \t
                • Take on strategic tasks<\/li>\n<\/ul>\n\n

                  For those starting their careers<\/h3>\n
                    \n \t
                  • Focus on skills that AI cannot replace<\/li>\n \t
                  • Communication, judgment, responsibility<\/li>\n<\/ul>\n\n

                    For non-academics<\/h3>\n
                      \n \t
                    • Craftsmanship and physical labour remain stable<\/li>\n \t
                    • High future security in practical professions<\/li>\n<\/ul>\n\n

                      For career changers<\/h3>\n
                        \n \t
                      • Choosing professions with social interaction<\/li>\n \t
                      • Mastering AI as a tool<\/li>\n<\/ul>\n\n

                        Conclusion<\/h2>\nThe Anthropic study doesn't show an extreme scenario \u2013 but rather a nuanced reality.\n\nAI won't suddenly replace millions of jobs. Instead, it will change them:\n
                          \n \t
                        • Access to the labour market<\/li>\n \t
                        • Productivity requirements<\/li>\n \t
                        • Value creation within professions<\/li>\n<\/ul>\n\n
                          The crucial question isn't: \"Will I lose my job?\" \u2013 but rather: \"Am I using AI better than others?\"<\/div>\n\nThe revolution is happening \u2013 quietly, but profoundly.\n\n