Microsoft study reveals the 40 jobs most at risk by AI and the 40 most secure

This reports provides a comprehensive overview of Microsoft’s recent study, “Working with AI”, published in July 2025, which examines the impact of generative AI on various professions. The study, conducted by Microsoft Research, analyzed 200,000 anonymized Bing Copilot conversations, mapping user goals and AI actions to the O*NET database, which categorizes U.S. jobs by their core activities. The findings, relevant as of July 28, 2025, offer insights into which jobs are most and least affected by AI, with implications for workforce transformation. This analysis is based on publicly available information and aims to present a detailed, professional perspective for readers interested in AI’s labor market effects.

The study, titled “Working with AI”, was published on arXiv with the identifier 2507.07935, accessible at https://arxiv.org/abs/2507.07935 and https://arxiv.org/pdf/2507.07935.pdf. It builds on previous research by analyzing real-world AI interactions, focusing on how generative AI, particularly through Microsoft Bing Copilot, interacts with human work. The methodology involves a crucial distinction between what users want to accomplish (user goals) and what the AI actually does (AI actions). This distinction revealed that in 40% of cases, the user’s goal and the AI’s activity involved different sets of tasks, highlighting AI’s role as a coach or advisor rather than a direct replacement.The researchers developed an “AI Applicability Score” to quantify AI’s potential influence, considering factors like frequency of AI use, success rate, and the completeness with which AI can handle tasks. This score was mapped to the O*NET database, ensuring a comprehensive classification of U.S. jobs. The study is specific to Microsoft Copilot usage in the U.S., with noted limitations for other platforms, countries, and informal work, such as household labor.

The evidence leans toward AI having a significant impact on knowledge-based roles, particularly those involving information gathering, writing/editing, and customer communication. Conversely, physical, hands-on jobs remain largely unaffected due to minimal overlap with AI’s current capabilities. Below are the detailed findings:

• AI’s Role: In 40% of conversations, AI performed different work activities than intended, suggesting it acts as a supportive tool rather than a direct replacement. This aligns with its strengths in assisting rather than automating entire roles, particularly in knowledge work.
• Correlation with Wages and Education: There is little correlation between AI suitability and pay or education level, though jobs requiring a bachelor’s degree are slightly more affected. This indicates AI’s impact is not strictly tied to traditional markers of job value.
• AI’s Strengths and Weaknesses: AI excels at tasks like collecting information, writing/editing, and communicating ideas, which are common in the most affected jobs. However, it is less effective for data analysis, visual design, and physical tasks, which are prevalent in the least affected jobs.

Lists of Jobs Most and Least Affected by AIThe study provides detailed lists based on AI applicability scores, with Table 3 listing the top 40 occupations most affected and Table 4 listing the bottom 40 least affected. Below are the complete lists, including coverage, completion rate, scope, score, and employment figures where available:

Table 1: 40 Jobs Most Affected by AI (High AI Applicability Score, Table 3)These occupations, primarily knowledge-based, have scores indicating significant AI impact, with tasks like writing and communication being highly automatable.

Table 2: 40 Jobs Least Affected by AI (Low AI Applicability Score, Table 4)These occupations, characterized by physical work with people, operating or monitoring machinery, and manual labor, have scores under 0.03, indicating minimal AI impact.

There is some controversy around AI’s impact on jobs, with debates on whether it will create new roles or displace existing ones. The study notes that while AI may automate certain tasks, it also has the potential to enhance productivity and create new industries, particularly in areas like AI development and human-AI collaboration. However, reports from the UN and World Economic Forum suggest AI could affect 40% of global jobs, potentially widening inequality, especially in developing nations absent from major AI governance discussions. This highlights the need for reskilling and upskilling initiatives to mitigate impacts.Microsoft’s own actions, including layoffs of over 15,000 employees in 2025 amid AI investments, reflect the practical implications of these findings. CEO Satya Nadella has acknowledged the “messy but necessary” transformation, emphasizing AI’s role in redefining work, which aligns with the study’s findings but also fuels debates on job security.

The study’s limitations include its focus on U.S.-based Microsoft Copilot usage, which may not generalize to other platforms or countries. It also excludes informal work and household labor, potentially underrepresenting certain sectors. Future research could explore AI-robotics integration, as noted in related articles, which may impact physical jobs like truck driving, currently listed as least affected.This detailed analysis aims to provide a comprehensive understanding of Microsoft’s study, offering insights for policymakers, workers, and businesses navigating the evolving landscape of AI and employment as of July 28, 2025.