Your next assignment could come from an AI manager that sets deadlines and evaluates your performance. In “The future of the firm,” OpenAI researchers Hemanth Asirvatham and Elliott Mokski explore how AI could eventually supervise employees and advise executives.
Automating coordination could give employees more control over their daily work by reducing reporting and approval steps. But if executives use AI to monitor performance and recommend personnel decisions, the same technology could concentrate authority at the top.
Why AI could take over management tasks before frontline jobs
Asirvatham and Mokski describe AI as an assistant or colleague. Another possibility places it between executives and employees. Some management responsibilities could be automated before frontline jobs because coordination and information analysis happen largely through digital systems, while other work requires physical activity or specialized tools.
The researchers acknowledge the uncertainty: “We don't yet know how good the models are at management.” The essay explores possibilities, rather than demonstrating effective employee supervision.
Middle managers who collect status updates and coordinate assignments could see those responsibilities shift to AI. Employers might give each remaining manager a larger team or reduce management positions.
Human managers would still need to resolve disputes and coach employees. Managing more people, however, could leave less time for individual development and difficult personnel conversations. Those demands could consume the hours automation saves.
How an AI manager could influence employee evaluations
In the essay’s hypothetical product launch, an AI adviser discovers through Slack that sales promised a feature engineering isn’t developing. It recommends a delay and reassignment of engineers, and identifies an overlooked contributor.
Such recommendations depend on how accurately AI interprets workplace communications. An informal suggestion could be mistaken for a customer commitment, while a conversation outside Slack might explain an apparent project delay.
AI could help executives recognize employees whose contributions rarely appear in formal reports. Yet a system that relies on searchable records might favor documented output over less visible work, such as mentoring a colleague.
Employees may also become less willing to discuss mistakes or share unfinished ideas if those conversations could influence their ratings. Employers would need to explain which communications inform evaluations and let workers challenge the system’s interpretation.
What workers could lose when managers delegate to AI
A 2026 study by researchers at Carnegie Mellon and Emory examined how employees might respond when AI communicates on behalf of human supervisors. Across six speculative workshops involving 23 participants, people expressed concerns about workplace surveillance and unclear accountability. They also questioned whether delegating to AI could weaken relationships with managers.
Although the workshops explored hypothetical situations rather than actual deployments, participants identified a potential cost: routine manager interactions provide feedback and opportunities to raise concerns.
Junior employees could face a related problem if AI takes over both routine assignments and interactions that help them learn. Reviewing an agent’s work requires judgment that often develops through supervised practice. Employers would need to preserve those learning opportunities so workers can develop the expertise to evaluate AI output.
What existing research says about supervising AI
Anthropic’s October employment research offers an example of employees supervising growing volumes of AI work. By August, employee-directed agents accumulated six times the estimated human work hours. The comparison uses agent runtime and assumes a forty-hour human workweek.
The report also found that junior employees’ share of technical hires fell from 27% to 9%, while employees increasingly developed tools to supervise agent output. These findings document changing responsibilities at Anthropic. They do not establish that businesses are generally replacing human managers.
In Microsoft’s 2026 Work Trend Index, 86% of 20,000 surveyed AI-using workers across ten countries said they treated AI output as a starting point rather than a final answer. Such attitudes could inform how companies introduce AI-assisted management, although the survey does not establish whether respondents identify mistakes.
Together, these findings suggest a potential complication: employees may supervise AI-generated work while receiving instructions or evaluations from AI systems themselves. Challenging a system’s output becomes more difficult when the recommendation carries a supervisor’s authority.
Who is accountable for an AI manager’s decisions?
Before using AI recommendations for promotions or disciplinary action, employers would need to identify who makes the final decision and how employees can challenge it. Workers should be able to examine the evidence behind an assessment and correct inaccurate records. A reviewer with authority to change the outcome should handle disputes.
AI could allow executives to oversee more employees with fewer intermediary managers. But fewer reporting levels would not necessarily give workers more control. Senior leaders could use automated reports to intervene more frequently in individual assignments and evaluations.
Employee independence would depend on which decisions companies leave with workers and when executives can intervene. Automating reports may remove admin work, but transferring authority requires an explicit management decision. Whoever approves an AI-assisted personnel decision should remain responsible for explaining it and correcting errors.
