23 September 2025 Generative AI and the workforce: More redistribution than reduction AI is reshaping the workforce through task redistribution rather than job cuts, with effects differing by industry, company size, and function. By Alex Johnston, Pollyanna De Lima, Sophie Malin, and Geoff Fitzgerald This is a thought leadership report issued by S&P Global. This report does not constitute a rating action, neither was it discussed by a rating committee. Highlights Department heads across various industries and functions surveyed in late 2024 and early 2025 projected that AI investment would have a flat to positive overall impact on employment. However, these respondents anticipated redistribution of tasks and shifting employment patterns. The employment impact of AI varies by company size and sector, with larger firms more likely to plan head count reductions. Roles in large corporations are typically more specialized and therefore easier to automate compared to the multi-functional positions common in small and medium-sized enterprises. The IT function in enterprises has seen the strongest negative employment implications related to AI investment and task automation, alongside administrative roles. These areas are often governed by clear productivity and output metrics, making it easier for organizations to justify and scale automation initiatives. Key drivers accelerating AI-powered task automation include an early strategic focus on cost reduction and operational efficiency, coupled with substantial investment in AI agents. Decelerators include concerns over security and governance, quality of AI outputs and employee resistance. The relationship between technology and automation has long been a subject of debate, one brought into sharp focus with the emergence of generative AI. This technological shift, which gained widespread public attention following the release of ChatGPT in November 2022, is often framed as a new frontier for automation. Unlike earlier, rules-based automation systems that focused on routine and simple tasks with pre-set flows, generative AI extends automation into more creative and challenging domains. This stems from the greater diversity of tasks to which generative AI models can be applied, with models able to generalize across domains due to the huge corpus of training data. This report is the first in a two-part series. A follow-up, scheduled for early 2026, will revisit organizational forecasts and evaluate the extent to which predictions made in 2024 materialized in 2025. How many jobs were forecast to be lost in the first wave of generative AI investments? Between August and September 2024, S&P Global Market Intelligence surveyed 766 department heads across sales, supply chain, IT, finance and accounting, and human resources functions in the energy and utilities, retail and manufacturing sectors in the US. Respondents projected that within a year, their workforce needs would decline by a median of 7.5%, and the average duration of assessed processes would decrease by 6.9%. However, these efficiency gains were not expected to translate entirely into job losses. In many cases, these figures were anticipated to come from reducing new hires, eliminating positions as employees voluntarily leave, or reallocating employees to new tasks. Nonetheless, this was expected to affect employment, with job growth slowing further into 2025, which will translate into higher sector or total unemployment levels. We're not cutting people … We are cutting the number of jobs we were planning to hire. We are not laying off people. We are just decreasing the number of open positions … So, instead of recruiting 100 people, we are just starting with 30, and we believe we are going to go up no further than 50 jobs instead of 100. Chief HR officer, utilities/energy, 1,000-4,999 employees, North America Source: 451 Research in-depth interviews, September 2024. I would expect it to be, at least, 10%-15% time savings … in my case it would provide me the opportunity to focus on other projects and or tasks that I may be delayed in or maybe never get to… the real savings are that I can achieve more. General manager of HR, retail, 500-999 employees, US Source: 451 Research in-depth interviews, September 2024. The S&P Global Market Intelligence Purchasing Managers' Index™ (PMI®) special survey on AI, conducted in October 2024, also reflected this theme of job reallocation, rather than reduction. Global organizations revealed a noteworthy outcome: a balance indicating zero net impact on employment from current AI investment. This net balance was calculated by deducting the percentage of survey respondents reporting lower and slightly lower head count due to AI investments over the past 12 months from the percentage reporting higher and slightly higher head count. While qualitative feedback suggested reduced requirements for positions in administration, marketing, sales and quality control, these decreases were offset by the creation of new roles. Employment changes linked to future AI investment for the 12 months to October 2025 were anticipated to be slightly positive, reflected in a net balance of +5%. The PMI AI survey showed that expected employment effects differed by company size. AI investment was projected to increase staffing at small (+7% net balance) and medium-sized enterprises (+11% net balance), whereas large corporations were more likely to anticipate a decline in employment, demonstrated by a net balance of -4%. This negative outlook among large firms stemmed from expected workforce reductions in the US, Germany and France. The greater automation risk for roles in large corporations partly reflects the fact that SMEs often employ individuals in multi-functional roles, making full role elimination less likely since these employees can shift to other core tasks. It also relates to differences in investment maturity. In 451 Research’s Voice of the Enterprise: AI & Machine Learning, Use Cases survey, larger organizations consistently report higher levels of AI investment and adoption maturity compared to those in lower revenue and head count tiers. Additionally, the return on investment from automation tends to be significantly higher at scale, and thus shapes project objectives and priorities. For example, while only 17% of respondents from organizations with fewer than 1,000 employees explicitly track head count reduction as a metric in AI initiatives, this share is nearly five percentage points higher among organizations with more than 1,000 employees. The trend toward longer-term reallocation rather than replacement was also reflected in the S&P Global Market Intelligence survey of department heads referenced above, with 32% expecting their departmental head count to remain stable over the next three years while 30% forecast an increase. The company's philosophy is to always increase head count. It's all about the customer and front facing with the customer … where there could be potential head count concerns is more on the manufacturing side because of what we do and how we build it … So, a lot of that has just been automated already through our own products, our own software and robotic manufacturing. Senior director of business development, manufacturing, 1,000-5,000 employees, US Source: 451 Research in-depth interview, September 2024. However, responses from functional heads suggest some roles and departments are more exposed to automation. For example, IT department heads projected that their head count needs for existing work would fall by a median of 9.8% over the 12 months to October 2025. Many acknowledged that generative AI would have significantly more impact on process time and head count than other forms of AI investment, such as pattern-recognition or rules-based models. Software development has emerged as a key application area for generative AI, and this may be reflected in the larger head count and process time reductions anticipated by IT compared to other departments. This emphasis is influenced by generative AI efficiency gains: reducing the time spent searching for, assembling and editing code. Further, a longstanding emphasis on tracking productivity metrics in software development has made it easier to quantify and consequently justify the value of AI investments in IT. As a result, organizations often find it simpler to move from pilot projects to full-scale deployment in this domain in comparison to areas with less-tangible objectives, such as improving employee experience. Industries expressed diverse views on how AI would influence their hiring strategies (see figure below). In the food and drink manufacturing sector, the proportion of businesses anticipating workforce growth due to AI adoption was 20 percentage points higher than the proportion expecting reductions. In contrast, the transportation manufacturing industry reflected a more cautious outlook, with the proportion of firms expecting to reduce head count being 10 percentage points higher than the proportion expecting to add staff. Thus, the size and sector of business deploying AI appears to have clear implications for the direction and scale of employment impacts, likely driven by the intentions behind the AI implementation as well as by the way in which AI-driven task restructuring flows through the business. We do not foresee AI negatively impacting the current capacity levels because … I don’t think we are at a place where we can simply use GitHub Copilot to write the code and then rely on it to write, to do integrations, to do new ETL code and new ingestions of data … There’s an aspect of human analysis that must happen. Head of data and analytics, manufacturing, >10,000 employees, US Source: 451 Research in-depth interview, September 2024. What tasks are being automated? There has been considerable speculation about whether generative AI and agentic AI could automate more strategic or specialist knowledge-based roles — positions that were largely unaffected by earlier waves of rules-based automation. However, this narrative is not clearly reflected in our data. Instead, the findings indicate that long-standing targets of automation, particularly administrative operations, remain the most exposed. While respondents pointed to data and information management as key areas for cost reduction, administrative operations were identified as the most automatable in terms of head count impact. The impact of technological trends on the automating potential of generative AI A range of technological advancements may contribute to greater displacement of human labor in current workflows over the coming year. Possible accelerators Increased investment In the October 2024 Purchasing Managers' Index™ (PMI®) Business Outlook surveys, 15% of global private sector firms said they had committed resources to AI projects, while 23% planned to do so in the next 12 months. More effective AI operationalization Organizations are moving beyond the initial wave of generative AI investments and adopting more disciplined approaches. This shift includes deeper integration of AI into core IT and business strategies, alongside structured governance, knowledge sharing and cross-functional collaboration. Centralized oversight models, such as chief AI offices and centers of excellence, are emerging to balance localized innovation with enterprise-wide standards, placing greater emphasis on data quality and governance. These developments provide a stronger foundation for automation than the commonly untargeted spread of exploratory investments which characterized early generative AI adoption. It's a little bit of a hybrid approach. So, there is a central chief AI office … that essentially acts as the gatekeeper for the AI/ML enterprise platform, so that the LLMs and the security posture, all of that is centrally controlled so that people are not doing random things. But … each line of business is free to spin up their own little AI tech team. VP of AI/ML engineering, healthcare/life sciences, >50,000 employees, US Source: 451 Research in-depth interview, April 2025. The drive toward AI agents While AI assistants or copilots typically augment or enhance human work, AI agents generally automate it, performing tasks independently, without direct human intervention. It is mostly internally facing. So, the use of AI is supporting colleagues in doing things faster. But it's also externally facing in some limited cases with our current architecture, and actually able to execute complete full processes … (our) new … chatbot-as-a-service platform aims to be able to entirely complete business processes on behalf of customers through agents. Lead enterprise architect, financial services, >50,000 employees, UK Source: 451 Research in-depth interview, April 2025. There is a clear desire to engage with more AI agents. Respondents to 451 Research's Voice of the Enterprise: AI & Machine Learning, Use Cases 2025 survey expressed widespread interest in agents, with 98% indicating interest and 58% actively seeking opportunities to implement agents or assistants in their organization. Cost and efficiency focus Organizations commonly seek efficiency and cost savings through their AI investments. While organizations may not explicitly include head count reduction as a success metric, it may be a natural follow-on. In the 451 Research survey referenced above, organizations identified cost reduction, process efficiency and employee productivity as the three most common KPIs for AI initiatives after project cost. Respondents cited data entry and analysis (53%), task automation (52%) and process optimization (51%) as the three main application areas for AI agent or assistant investments. Among the various sectors responding to the PMI AI survey, productivity enhancement was the top-cited benefit of AI investment. In nine of the 15 sectors surveyed, more than 50% of respondents expected productivity gains. Even in the hotels and restaurants sector, which is traditionally associated with automation-resistant, face-to-face service, 39% of companies preparing to invest in AI anticipated productivity gains. If we're going to do AI, it needs to be cost efficient or offset by savings that we can justify. Otherwise, it won't go off the ground. IT/engineering manager/staff, healthcare, 5,000-9,999 employees, revenue unrecorded Source: 451 Research’s Voice of the Enterprise in-depth interview, May 2025. Possible decelerators Ongoing concerns related to data privacy, governance and security Several factors hold back enterprise adoption of generative AI. As the figure below illustrates, organizations face major challenges with data privacy and security. These concerns become especially acute when organizations move beyond low-risk use cases, such as generating generic marketing content, and begin exploring data-intensive or mission-critical applications that rely on proprietary data. Agentic workflows bring additional layers of orchestration complexity, which can also increase risk by opening up a larger attack surface or via non-malicious, unintended outcomes such as agents exploiting legal loopholes to achieve results in the absence of explicit ethical guidelines, or pursuing short-term goals over long-term value alignment. Privacy risks extend beyond simply exposing sensitive information to third-party AI providers. Enterprises must also ensure that generative AI tools do not inadvertently grant employees access to data they are not authorized to view. [Our IT] priority is getting a handle on how we can leverage AI safely. Zero trust is a huge one, but real zero trust, not the marketing hype that everybody talks about. Mid-level management, business services, 250-499 employees, $2.5M-$4.9M revenue Source: 451 Research’s Voice of the Enterprise in-depth interview, June 2025. Many organizations see a need to implement robust governance frameworks and continuous monitoring before meaningfully progressing with their AI initiatives. In the same 451 Research AI & Machine Learning, Use Cases study cited above, just 27% of organizations said they had invested in MLOps tools, platforms or functionality, with a further 42% planning to in the next 12 months. Some organizations also see a need to address the ethical considerations of AI, delaying projects until they have developed policies governing responsible and fair use. So clearly, regulatory compliance, especially as we are in the public sector. So, this one is compulsory. Chief information and technology officer, government, 10,000-20,000 employees, France Source: 451 Research in-depth interview, April 2025. Infrastructure limitations More than half of respondents (54%) to 451 Research's Voice of the Enterprise: AI & Machine Learning, Infrastructure 2025 survey said they required additional AI accelerators in the cloud for AI training or inference to improve the performance of their AI initiatives. A further 46% were seeking on-premises graphics processing unit (GPU) servers. Other commonly identified infrastructure gaps included storage, networking and memory capacity. The growth of generative AI is limited not just by these infrastructural bottlenecks, but also the energy required to support them. AI output quality Quality concerns are emerging as a significant barrier to generative AI adoption, with 29% of organizations reporting that the accuracy of outputs is challenging progress. This issue has grown year over year and is most pronounced among higher-maturity organizations. Trust in AI predictions has notably declined. Only 24% of respondents to 451 Research’s Voice of the Enterprise: AI & Machine Learning, Use Cases 2025 survey say they have complete confidence in AI outputs, down from 2023 levels as awareness of generative AI’s propensity for hallucination has increased. A key driver of this challenge is the inability of current models to adequately integrate semantic understanding, procedural reasoning and user-specific long-term memory, with many use-cases requiring a level of domain and contextual awareness that general-purpose large external models cannot natively deliver. Enhancing accuracy of model outputs is a key requisite to drive higher adoption of agentic AI in workflows, as it will enable gradual trust buildup. It is [yet] to be proven that it works without negative impacts, like the hallucinations or toxicity, or based on data quality. So, we have to have those guardrails built-in and human oversight before the user is leveraging or taking an action that might negatively impact them. Head of data & analytics, manufacturing, >10,000 employees, US Source: 451 Research in-depth interview, September 2024. Employee resistance Over a quarter of respondents to 451 Research's Voice of the Enterprise: AI & Machine Learning, Use Cases 2025 said their organizations were concerned about staff resistance (28%) or customer resistance (27%) to their AI initiatives. These concerns were among the top five issues organizations identified, and only narrowly surpassed by confidence in accuracy (29%) and budget limitations (29%). This resistance may pose a challenge because inadequate levels of usage, limited skill at using AI-enabled tools or union demands to limit unemployment risks may undermine the value of investments. Change management will be big. If you don't get people believing in it and wanting to do it, it's very hard to get it done. And oftentimes the people you need to help you build the models are the people that are impacted by the AI tool that you're bringing in … that's why we're all constantly talking about, ‘Hey, this is to help you, not to replace you,’ right? VP of supply chain and inventory, retail, >10,000 employees, US Source: 451 Research in-depth interview, September 2024. Implications In S&P Global Market Intelligence surveys conducted in 2024 and early 2025, organizations broadly expressed expectations that AI would reshape existing work by boosting efficiency and redistributing tasks, rather than causing widespread job losses. However, the projected impact on head count differed by sector and company size. IT and administrative roles faced the greatest potential impact. It remains to be seen how significantly concerns over data privacy, governance and employee resistance will act as decelerators, slowing the likely pace of transformation. A report in the first quarter of 2026 will seek to validate these forecasts, tracking how attitudes to automation are evolving alongside the technology. Explore more research on Artificial Intelligence from S&P Global Read More Contributors: Matt Tompkins, Mahnoor Haider, and Paul Whitfield