A recent eight-month field study at a 200-person U.S. tech company found that AI work expanded employees’ responsibilities, pushed work into breaks and evenings, and increased multitasking. The authors treat that pattern as a warning. For HR leaders, the more useful conclusion is sharper: AI does what serious productivity tools usually do. It raises capacity, then the organization quietly rewrites the job around that new capacity.
You should not read the result as surprising. Acemoglu and Restrepo’s automation and new tasks framework describes how technology shifts tasks away from humans in some areas while creating or enlarging tasks in others. Generative AI accelerates that pattern inside knowledge work because it strips friction from drafting, summarizing, research, coordination, and routine analysis. Once those tasks take less time, managers and employees rarely bank the time as slack. They move attention toward judgment, cross-functional execution, quality control, and faster decision cycles.
AI Expands the Job Before It Eliminates the Job
HR should treat AI as a job architecture event, rather than a software rollout. In one real-world study of a generative AI assistant used by 5,172 customer support agents, researchers found AI productivity gains averaging about 15%, with the largest benefits for less experienced and lower-skilled workers. That finding matters for HR because it shows two truths at once: AI can improve performance, and it can change who learns from whom.
When AI absorbs the monotonous layer, human work moves upward. Analysts become translators between functions. Product managers draft artifacts that previously waited in a technical queue. Engineers spend less time on boilerplate and more time reviewing AI-enabled work, setting standards, and teaching the organization what good looks like. Employees who navigate the transition well can gain more autonomy and stronger job security because they own outcomes rather than chores.
That same move upward creates strain. If a product manager can produce a rough prototype, someone still needs to review the code, assess risk, and decide whether the work belongs in production. If a recruiter can generate a role scorecard faster, someone still needs to test whether it reflects the real job and avoids wishful thinking. AI shortens the first draft. It does not remove accountability. That shift turns job descriptions into living documents, and HR should revise hem around outcomes, review responsibilities, and decision rights.
HR Must Turn Productivity Gains into Operating Norms
The HR risk does not come from AI use itself. It comes from using AI without a new operating system. An analysis of 164,000 workers across 443 million work hours found workload creep among AI users, including more time in communication tools and less uninterrupted focus time. The Berkeley field study points in the same direction, with task expansion, boundary creep, and multitasking. For HR, these are work design issues rather than wellness slogans.
Existing technostress research already gives HR a useful vocabulary. A validated technostress creators model includes overload, invasion, complexity, insecurity, and uncertainty, which map cleanly onto current AI adoption. AI can make work faster, blur work boundaries, force constant learning, raise fear about job security, and keep employees chasing changing tools. When leaders celebrate the throughput while ignoring those conditions, they turn a productivity gain into a retention risk.
This does not mean HR should slow AI adoption. It means HR should set explicit norms before informal behavior hardens. Teams need rules for when AI work can run in the background, when people can start tasks during breaks, how much AI-generated work engineers or managers should review, and which decisions require human signoff. Managers need capacity planning that counts review work, rather than only production work. Employees need permission to stop, rather than merely encouragement to automate.
Entry-Level Roles Need a New Apprenticeship Model
The deepest HR problem sits in talent development. Generative AI performs many starter tasks: first drafts, basic research, data cleanup, ticket triage, routine reporting, and preliminary analysis. Those tasks frustrated new employees, but they taught context, standards, judgment, and organizational language. They also justified hiring people before they had much experience.
Stanford’s Digital Economy Lab found an entry-level jobs warning signal in AI-exposed occupations: workers ages 22 to 25 experienced a 16% relative employment decline after widespread generative AI adoption, while less exposed fields and more experienced workers remained stable or grew. The same Stanford paper found declines concentrated where AI automates rather than augments human labor. For HR, that distinction should govern workforce planning.
If companies erase the lower rungs, they should expect a future shortage of people ready for higher rungs. Hiring only experienced workers can work for one firm briefly. It cannot work for the labor market as a whole. A company that uses AI to eliminate routine work must rebuild the learning function that routine work used to provide.
That requires a different talent pipeline strategy. New hires should rotate through supervised AI-first workstreams, where they prompt, check, explain, and improve outputs under explicit review. Early-career employees should practice comparing AI drafts against source materials, identifying errors, asking better questions, and translating output into business decisions. Mentors should assign judgment reps, rather than busy work. Apprenticeship should become more deliberate because AI removed some accidental learning.
In evaluating AI adoption at work, the organizations that improve fastest treat AI as a change in work design rather than a pile of licenses. The same logic applies to flexible work norms because AI can improve autonomy only when leaders define expectations around availability, boundaries, and results.
AI will make many jobs bigger before it makes them smaller. HR’s job is to ensure bigger means smarter, more developmental, and more sustainable, rather than merely faster.
Gleb Tsipursky, PhD, serves as the CEO of the future-of-work work consultancy Disaster Avoidance Experts and wrote The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026) and Returning to the Office and Leading Hybrid and Remote Teams (Intentional Insights, 2021)