AI and Team Size: Is the Team of Five Now One Person and an AI?

AI and team size are linked, but not in the way the headlines suggest: the best evidence shows that AI raises productivity on specific tasks, often by 15–40%, yet it has not turned teams of five into one person and an AI. Where teams are getting smaller, the main channel is fewer hires, especially of junior staff, not mass layoffs.

That distinction matters for HR. A team that shrinks because nobody is hired to replace leavers or to fill junior seats looks very different, in workforce plans, pay structures and career paths, from a team that is cut. This article reviews what peer-reviewed and official research actually shows, adds TalentUp’s global job-ad data for 2024–2026 and turns it into concrete steps for HR and compensation and benefits (C&B) teams.

Is AI Shrinking Teams? The Short Answer

The idea that “the team of five is now one person and an AI” comes mostly from start-up anecdotes and vendor marketing. The research tells a more measured story:

• Task level: AI makes people faster and, for less experienced workers, better at a defined set of tasks. On tasks outside AI’s capabilities, it can make them worse.
• Firm level: companies that adopt generative AI are hiring fewer junior workers than comparable companies, while senior employment keeps growing.
• Economy level: large studies using payroll and administrative data find no broad fall in employment, hours or earnings so far, with one important exception: young workers in occupations where AI automates tasks.

In other words, AI is changing the shape of teams (fewer entry points, more senior work per person) faster than it is changing their size. That is still a major workforce planning problem, just a different one from the one most headlines describe.

What the Research Shows About AI Productivity

The strongest evidence comes from randomized experiments and field studies with real workers. These are the studies most often cited, with what they actually found:

The Procter & Gamble experiment is the closest to the “one person and an AI” claim, and it is worth reading carefully. One professional using AI performed as well as a team of two without AI on a specific innovation task. Teams with AI still did best. It is evidence that AI can substitute for some of the benefits of collaboration, not that it can replace a five-person team.

Where AI Gains Are Large and Where They Are Small

The consultants’ study coined the term “jagged technological frontier”: AI is very good at some tasks and surprisingly poor at others that look similar. That explains why the numbers above range from a 40% gain to a 19% loss.

Large-scale data confirms that lab gains do not automatically become labor-market changes. Humlum and Vestergaard (2025) linked AI chatbot adoption surveys to Danish administrative records for about 25,000 workers in 11 exposed occupations, and found effects on earnings and recorded hours that were close to zero in 2023–2024. Workers reported modest time savings of around 3%, and part of the saved time went to new tasks created by the AI itself, such as checking output.

For HR, the lesson is that headcount savings are not automatic. Productivity gains only change team size when work is redesigned, and they are concentrated in specific tasks, not whole jobs.

Hiring, Not Firing: The Entry-Level Squeeze

The clearest employment effect of AI so far is on people starting their careers.

• Stanford “Canaries in the Coal Mine” (Brynjolfsson, Chandar and Chen): using ADP payroll records through June 2026, employment for workers aged 22–25 in the most AI-exposed occupations is about 19% below where it would be had it kept pace with less exposed peers (15% in the July 2025 data). Experienced workers show no comparable gap. The authors find the adjustment works primarily through reduced hiring of young workers rather than increased separations, and that declines concentrate in occupations where AI automates rather than augments work. They describe these as descriptive patterns, not causal estimates.
• Hosseini and Lichtinger (Harvard, 2025): with résumé and job-posting data for about 62 million workers in 285,000 US firms, they find that from 2023 junior employment in firms adopting generative AI declined sharply relative to non-adopters, while senior employment continued to rise, again driven by slower hiring rather than more exits.

This is how a team gets smaller without anyone being fired: a senior leaves and is not replaced, or the two graduate seats planned for next year are quietly removed. We cover the European picture in detail in why entry-level jobs in Europe are becoming harder to find, and what this means for career-switchers in can a bootcamp prepare you.

The long-term risk for employers is a broken talent pipeline: if nobody hires juniors today, there will be fewer experienced people to promote or hire in five years.

What Job Ads Show: TalentUp Data for 2024–2026

TalentUp collects job ads worldwide for more than 800 roles. We looked at three questions: are employers asking for AI tools as a skill, are they still hiring juniors in AI-exposed roles, and are ads for the most exposed roles disappearing?

Source: TalentUp job postings database, job ads worldwide, January–September 2024 and 2026. See methodology below.

AI tools have become a job requirement

The shift is fastest in the roles where research finds the largest task-level gains: writing and software. Typical wording is “experience with AI-assisted development tools (e.g., GitHub Copilot)” or “comfort using AI tools to speed up workflows”. In customer support, most mentions are not about the job at all but about AI used in the hiring process, a sign that AI there is deployed by the company rather than handed to the agent. A further 4.1% of writing ads in 2026 are for AI training work, such as writing and evaluating content for language models, up from 2.9% in 2024.

Fewer junior seats in AI-exposed roles

The junior share fell in most knowledge-work families, and fastest in software development, where it dropped by more than a third in two years, against a much smaller decline across all ads. Marketing is the exception. This is consistent with the “hiring, not firing” pattern in the research: employers keep advertising developer roles (their share of all ads has actually grown) but ask for more experienced people.

Exposed roles are not disappearing from job ads

Ads for the roles most often described as “replaceable by AI” have not collapsed. Between 2024 and 2026, customer support roles went from 0.47% to 0.55% of all ads, data entry from 0.11% to 0.10%, and writing and content fell from 0.15% to 0.12%, with most of that fall between 2024 and 2025. Translator ads rose, mainly because of freelance platforms. Ads for AI and machine learning engineers grew from 0.14% to 0.53% of all ads. Changes in the mix of sources we collect from make these shares less reliable than the within-role comparisons above, so read them as “no collapse”, not as precise trends.

How Roles and Pay Change When Teams Get Smaller

If fewer people do more work with AI, do they get paid more? The evidence so far says not automatically.

• In the Stanford payroll data, adjustment shows up in employment, not in base pay: wages of young workers in exposed jobs have not fallen, but fewer of them are employed.
• In Denmark, Humlum and Vestergaard find that only a small share of AI productivity gains has reached workers’ earnings.
• In the customer support study, AI helped new agents reach the performance of experienced ones faster, which compresses the value of tenure rather than raising pay for everyone.

What does carry a premium is specialist AI work. In TalentUp’s 2026 job ads with an annual salary, the median for AI and machine learning engineers is €172,200 in the US vs €147,200 for software developers (+17%), €111,900 vs €88,100 in the UK (+27%), €103,900 vs €87,000 in Canada (+19%) and €62,400 vs €43,700 in Spain (+43%). We break this premium down by country, role and skill in the AI pay premium in Europe.

The likely direction for roles is clearer than for pay:

What HR and C&B Teams Should Do

How AI is reshaping the HR function itself is covered in how AI is rewriting the HR operating model. For the rest of the organization, these are the priorities.

1. Plan workforce by task, not by headcount ratio

Do not budget on the assumption that AI cuts teams by a fixed percentage. Map each role’s main tasks, mark which are automated, augmented or unaffected, and test the gains in a pilot before changing headcount. The evidence shows large gains on some tasks and losses on others.

2. Protect the junior pipeline deliberately

If you stop hiring juniors, you will have to buy experienced talent later at market rates. Keep a minimum share of entry-level roles, redesign them around tasks AI does poorly (client contact, judgment, cross-team work) and use AI as a training tool, which is where research finds the largest gains.

3. Update job architecture and levelling

When a role absorbs AI-augmented work, its scope, accountability and level may change. Review job descriptions, levelling criteria and career paths so that broader roles are graded consistently, and avoid creating new job titles for what is really the same job plus a tool.

4. Pay for scope and outcomes, not for “using AI”

Using AI tools is becoming a baseline expectation, not a skill that justifies a premium on its own. Reserve premiums for specialist AI roles and for roles whose scope genuinely grew, and benchmark them against the market with TalentUp’s salary data rather than one-off offers. Under the EU Pay Transparency Directive, which applies through national laws once transposed, pay differences must rest on objective, gender-neutral criteria, so document why an AI-augmented role sits in a higher band.

5. Reskill before you restructure

Train people on the tools and on the jagged frontier: when to trust AI output and when not to. The METR and BCG results show that misuse can cost more than it saves.

6. Consult workers and meet AI Act obligations

In the EU, AI systems used for recruitment, promotion, termination, task allocation or evaluating workers are high-risk under the AI Act. Following the Digital Omnibus on AI (Regulation (EU) 2026/1744), those obligations apply from 2 December 2027 instead of 2 August 2026. Before using such a system at work, employers must inform workers’ representatives and the affected workers. National information, consultation and co-determination rules, such as works council rights, can require more, and any restructuring that leads to dismissals triggers the usual collective redundancy procedures.

Conclusion

The team of five is not yet one person and an AI. What the evidence shows is quieter but just as important: AI makes people faster at specific tasks, firms respond by hiring fewer juniors, and job ads increasingly treat AI tools as a standard requirement. Pay has not followed productivity, except for specialist AI roles.

For HR and C&B teams, the job is to manage that shift deliberately: plan by task, keep a junior pipeline, update job architecture and price AI-augmented roles on evidence. See TalentUp plans to see how TalentUp’s market data can support those decisions, or read our methodology.

Sources

• TalentUp. Job postings database, job ads worldwide, January–September 2024, 2025 and 2026. Methodology.
• Brynjolfsson, E., Li, D. and Raymond, L. (2025). Generative AI at Work. Quarterly Journal of Economics, 140(2). Working paper version: arXiv 2304.11771.
• Noy, S. and Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654).
• Dell’Acqua, F. et al. (2023). Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. Harvard Business School Working Paper 24-013.
• Dell’Acqua, F. et al. (2025). The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise. NBER Working Paper 33641.
• Becker, J., Rush, N., Barnes, E. and Rein, D. (2025). Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. METR.
• Humlum, A. and Vestergaard, E. (2025). Large Language Models, Small Labor Market Effects. NBER Working Paper 33777 / BFI Working Paper 2025-56.
• Brynjolfsson, E., Chandar, B. and Chen, R. (2025, revised August 2026). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab.
• Hosseini, S. M. and Lichtinger, G. (2025). Generative AI as Seniority-Biased Technological Change: Evidence from U.S. Résumé and Job Posting Data. Working paper.
• European Union. Regulation (EU) 2024/1689 (Artificial Intelligence Act). EUR-Lex.
• European Parliament. Digital Omnibus on AI, Regulation (EU) 2026/1744. Legislative Train Schedule.
• European Union. Directive (EU) 2023/970 on pay transparency. EUR-Lex.