OpenAI Authors Map AI’s Possible Roles in Firms

OpenAI authors Hemanth Asirvatham and Elliott Mokski argue that AI could reorganize companies through three arrangements: beneath workers as assistants and agent teams, beside managers as colleagues, or between executives and employees coordinating work. In their view, these scenarios are already appearing in subtle ways, and firms may combine them as AI improves. Their essay is the second in a series on the next economy, hosted on a platform for independent voices exploring an AGI future; the authors say its views are their own, not OpenAI’s or their colleagues’.

In the “promoting” scenario, employees keep their positions while AI agents initially handle tasks such as drafting emails and memos. As models improve, agents could take on more work, coordinate with one another, and form miniature organizations managed by people. The authors illustrate this with Jace, an accountant whose agents handle spreadsheet work while he shifts toward communication, coordination, and creative decisions. His accounting team does not shrink in the example because company productivity and demand for accounting work grow, while executives still need the team’s expertise to use the agents.

The authors say this arrangement does not necessarily mean layoffs: employees could take on broader responsibilities as AI handles parts of their former jobs. They argue that the outcome depends on whether people continue adding value and whether consumers want increased output; if demand does not grow, firms could instead employ fewer people or reduce hours. They also identify costs for junior workers: AI could take over entry-level tasks that traditionally taught workplace skills, although junior employees might gain resources to pursue ideas and learn through broader projects. They expect adoption may begin with frontline employees who understand the work AI is being asked to do.

In the “interleaving” scenario, AI teams work alongside human managers and staff. The authors describe a furniture-company executive who assigns an AI marketing team a budget and a sofa to promote. Its agents divide campaign work, interview salespeople, check with the factory, survey customers, and draft visual concepts; the executive reviews a campaign and authorizes a photographer to reshoot images. The authors say broader AI capabilities may be harder to build but easier for executives to adopt directly, while narrower tools often require workers to retrain and reorganize processes. They caution that AI inserted beside individual workers could create many handoffs, and that this scenario can coexist with AI agents beneath employees without necessarily reducing total employment.

In the “layering” scenario, AI sits between managers and workers, with some employees reporting to an AI system. The essay calls this arrangement an “automation sandwich” and says less radical forms may already have begun. The article’s supplied excerpt ends as it starts an example about Asha, a chief executive, receiving an AI consultancy’s warning about a problem with an upcoming launch; it does not provide the rest of that example.

OpenAI authors Hemanth Asirvatham and Elliott Mokski argue that AI could reorganize companies through three arrangements: beneath workers as assistants and agent teams, beside managers as colleagues, or between executives and employees coordinating work. In their view, these scenarios are already appearing in subtle ways, and firms may combine them as AI improves. Their essay is the second in a series on the next economy, hosted on a platform for independent voices exploring an AGI future; the authors say its views are their own, not OpenAI’s or their colleagues’.

In the “promoting” scenario, employees keep their positions while AI agents initially handle tasks such as drafting emails and memos. As models improve, agents could take on more work, coordinate with one another, and form miniature organizations managed by people. The authors illustrate this with Jace, an accountant whose agents handle spreadsheet work while he shifts toward communication, coordination, and creative decisions. His accounting team does not shrink in the example because company productivity and demand for accounting work grow, while executives still need the team’s expertise to use the agents.

The authors say this arrangement does not necessarily mean layoffs: employees could take on broader responsibilities as AI handles parts of their former jobs. They argue that the outcome depends on whether people continue adding value and whether consumers want increased output; if demand does not grow, firms could instead employ fewer people or reduce hours. They also identify costs for junior workers: AI could take over entry-level tasks that traditionally taught workplace skills, although junior employees might gain resources to pursue ideas and learn through broader projects. They expect adoption may begin with frontline employees who understand the work AI is being asked to do.

In the “interleaving” scenario, AI teams work alongside human managers and staff. The authors describe a furniture-company executive who assigns an AI marketing team a budget and a sofa to promote. Its agents divide campaign work, interview salespeople, check with the factory, survey customers, and draft visual concepts; the executive reviews a campaign and authorizes a photographer to reshoot images. The authors say broader AI capabilities may be harder to build but easier for executives to adopt directly, while narrower tools often require workers to retrain and reorganize processes. They caution that AI inserted beside individual workers could create many handoffs, and that this scenario can coexist with AI agents beneath employees without necessarily reducing total employment.

In the “layering” scenario, AI sits between managers and workers, with some employees reporting to an AI system. The essay calls this arrangement an “automation sandwich” and says less radical forms may already have begun. The article’s supplied excerpt ends as it starts an example about Asha, a chief executive, receiving an AI consultancy’s warning about a problem with an upcoming launch; it does not provide the rest of that example.