Evidence at a glance
From Doing the Grunt Work to Managing Agents
The Future of the Firm, an opinion essay published on OpenAI’s website by Hemanth Asirvatham and Elliott Mokski, asks how companies might redistribute work as AI agents become part of the workforce. The authors recall that they began managing teams at age 24. On a traditional career ladder, people at that stage might still be taking meeting notes and drafting memos, but the authors say they were already directing AI agents that worked around the clock. They explicitly note that the essay reflects their views, not those of OpenAI or their colleagues.
That experience is not evidence of a general shift across companies, but it surfaces a management question early: if young employees no longer enter organizations by doing routine work, and instead begin by directing digital labor, the change is about more than who completes each task. Authority over work allocation, responsibility for checking results, and the way newcomers build judgment may all shift with it. The essay is not announcing that AI will simply replace jobs. Its focus is the transition in which people and AI work together.
Three Positions, Three Management Relationships
The authors use three positions on the org chart to describe possible changes. “Beneath” means agents act as assistants or subordinates to employees. At first, each person might have one or two agents handling small tasks such as sending emails or drafting memos. As trust grows, agents could take on more work and coordinate with one another. Human job titles may remain unchanged, while employees shift from execution toward setting goals, reviewing output, and managing a small digital team.
“Beside” imagines AI as a functional colleague, moving from individual tasks toward a broader scope of work. “Between” is different: AI sits between human levels, coordinating, summarizing, and monitoring work so that managers can see progress, risks, or contributions. These are illustrative scenarios, not a statistical account of how firms are organized today. Their key distinction is whether AI assists an individual, takes on a parallel function, or controls an organizational coordination channel.
The Broader the Capability, the Higher Deployment May Move
The adoption path in the essay depends on how much of a job AI can perform. Tools that handle only fragments of work are likely to be tried, adjusted, and integrated by employees who know the day-to-day process. If AI can take on an entire function end to end, senior leaders may be able to deploy it more directly. The authors frame this as an organizational learning problem: technology must learn how a firm works, while employees already understand its routines.
As a result, more capable agents may do more than speed up individual tasks. They may also change who has the power to initiate automation. Frontline employees could get tools and new management responsibilities first, while decision-making later shifts upward if AI can cover a complete function. The essay does not establish that this transfer of authority must happen. It does identify a causal possibility for leaders to watch: the broader the function AI can handle, the more adoption may move from local experimentation to a decision made at the top.
Higher Productivity Does Not Automatically Mean Fewer Jobs
The authors use a hypothetical figure to show that automation does not dictate a single outcome for the work left over. If AI can perform 90 percent of an employee’s former workload, one possibility is that the employee has only about 10 percent of that work remaining. Another is that the employee uses the freed capacity to take responsibility for goals roughly ten times larger. This is a thought experiment about task reallocation, not a measured productivity result or a forecast of job numbers. The outcome depends in part on whether the firm needs more output and whether people can add value to the expanded work.
The essay draws on a broader lesson from historical research on automation: automation can substitute for labor, but it can also complement workers, raise output, and change demand for labor. That history does not prove which org chart AI will produce, and it cannot guarantee that productivity gains will become higher wages. For leaders, the question should not stop at how much time agents save. They also need to ask whether that capacity creates new business or services, or simply allows the same work to be done with fewer people.
Coordination Is Also a Form of Oversight
The “between” scenario most clearly exposes the tension between efficiency and authority. AI that can summarize communications, flag risks, and identify contributions might help managers spot coordination failures sooner. The same capability could also give senior leaders a more direct way to observe and supervise employees. It narrows the distance between management and work on the ground, but it does not ensure that the resulting picture is fair, complete, or more respectful of employee autonomy.
There is also a cost to early-career development. If agents take over email, memo drafting, and other basic work, newcomers may lose a route to learning the business and building judgment through repetition. Agents still make mistakes, and supervising them requires expertise. When piloting agents, technical leaders should design output review, accountability for errors, and ways for new employees to gain experience that routine work once provided. A practical decision is to evaluate separately who performs a task and who has the authority to see, interpret, and approve it. Automation can reduce execution effort, but firms still need explicit oversight boundaries and a deliberate choice about whether productivity gains go toward growth, employee development, or cost reduction.