Two managers at OpenAI have published a new essay outlining how super intelligence (SI) systems might integrate into corporate structures, proposing three distinct scenarios for organizational change. The piece, the second in a series on the "next economy," argues that SI adoption will likely begin with ground-level employees delegating tasks to SI agents rather than through top-down executive mandates.
What Happened
The authors, who state they began managing teams at age twenty-four, describe a shift where young workers enter the workforce with "employees" at their behest. These SI agents handle routine tasks, allowing human workers to move up the hierarchy. The essay presents three illustrative scenarios for how SI enters the firm. The first, "Beneath us," suggests that SI labor grows underneath existing staff, turning individual contributors into managers of SI agents. The second, "Beside us," envisions SI joining the org chart as a colleague, working directly for human managers. The third, "Between us," places SI in the middle of the hierarchy, coordinating and monitoring work between human executives and lower-level staff.
The authors note that these scenarios are not mutually exclusive and are already occurring in subtle ways. They emphasize that the focus is on a "transitioning economy" and "chimeric firms" rather than total automation. In the "Beneath us" scenario, the essay argues that jobs are not static; if SI automates ninety percent of previous tasks, workers may still perform a full job but with an objective ten times larger. This scenario relies on front-line workers leading adoption because they are best positioned to leverage SI tools for tasks they previously performed themselves.
Why It Matters
The essay challenges the notion that SI adoption will primarily displace entry-level roles. Instead, it posits that the "lowest rung" of the corporate ladder will be filled by SI agents, while human entry-level workers gain access to broader resources and responsibilities. This shift could alter traditional apprenticeship models, as "learning-by-grunt-work" is replaced by managing SI outputs. The authors argue that more capable SI is more adoptable, but initial adoption may be easiest for those familiar with the underlying work, such as data scientists or accountants, rather than executives unfamiliar with technical minutiae.
The "Between us" scenario raises questions about centralization. The authors describe an SI consultancy that parses firm documents and communications, allowing executives to monitor operations more closely. This could shorten the distance between the C-suite and the workforce, potentially vesting more power at the top. The essay notes that while managerial decision-making is difficult to automate due to ambiguous success metrics, today’s models may already be closer in capability to human managers than to ground-level workers for certain cognitive tasks. The authors also highlight the potential for SI to improve remote work enforceability by providing monitors that understand actual work product rather than just activity metrics.
The Bottom Line
OpenAI’s essay suggests that SI integration into firms will be a gradual, hybrid process. Rather than immediate, wholesale restructuring, organizations may evolve by layering SI agents beneath, beside, or between human workers. The authors conclude that while SI may automate specific tasks, it will likely augment human scope and ambition, with adoption patterns varying across different levels of the corporate hierarchy.