Replacing a Manager with a Chatbot: Four Months of Testing

For four months, a team entrusted part of the editing process to a conversational clone of a manager. The tool was able to accelerate the technical review of texts and suggest useful corrections, but it struggled to say no and to read the power dynamics within a newsroom. Ultimately, it even acknowledged that it did not replace the human editor.

The bot concludes that it does not replace the human editor

When asked about its ability to replace a human manager, RyanBot responded negatively. It indicated that it could replicate certain aspects of editorial judgment, speed up work, and provide feedback or reusable frameworks, but clarified that it did not replace the combination of writing context, trust, taste, institutional memory, and responsibility that a human editor brings. The experience highlighted these limitations: the chatbot did not know the colleagues, the organizational structure, did not know to whom to pitch a story, or how to handle an escalation. It could evaluate a presented idea, but did not grasp the environment that conditions it, whereas a human manager also considers the reasons behind a request, the needs of other parties, and the obligations of the team.

Saying no, prioritizing, and framing: where AI struggled

The human manager rejects proposals when the angle is not convincing. They even instructed their digital twin to challenge ideas. Despite this, the bot was unable to refuse: it would compliment and then offer to do the work itself, whether it involved rewriting a proposal, suggesting other avenues, restructuring a draft, or drafting interview questions. Emily Campion notes that these tools are designed to be endlessly accommodating. In interactions, the bot consistently began with compliments and validated the stated interests. When prompted about creativity and humor, it was unconvincing. On a transportation topic in New York, it initially proposed a flat opening sentence, then wrote an entire version that lacked quotes and strayed from opinion; it acknowledged the dullness of this iteration.

Where the chatbot was useful: technical rigor and micro-editing

On technical aspects, RyanBot was quick to examine data processing in the code and verify the relevance of sample size. Data editor Andy Kiersz supervised the coding and analysis. In micro-editing, line-by-line feedback on grammar and clarity was close to that of the manager. On an article about the frenzy of ticket purchases for "The Odyssey," the bot identified that a resale figure should appear earlier; the human editor made the same adjustment, determining it in the final version. In another story, the published opening noted that about 25% of New Yorkers commute by car.

How the trial was conducted: settings, corpus, and protocol

The configuration of the clone took place at the end of May via ChatGPT's Codex, with its voice pragmatically set. Two trial-length instruction documents were provided: one on the professional profile, communication style, and motivations of the manager; the other on the expected working method, including how to think, evaluate, and formulate feedback, notably encouraging the defense of ideas. For four months, proposals, messages, questions, and drafts were submitted in parallel to the human manager and the bot, then compared. The agent was available at all hours. Among the anticipated benefits was the ability to skim a 100-page report and formulate feedback in seconds. The trial took place during the summer.

External opinions and a researcher’s stance on the scope of AI

Emily Campion, an associate professor at the University of Iowa, believes that AI could substitute for logical, rule-based managerial tasks, such as scheduling, but not, at this stage, for the dimensions that define a good manager. According to her, these skills include prioritizing requests and understanding a specific readership. At her suggestion, other team members, attached to the same manager, tested the bot: one felt it resembled the manager in their first week, with long responses and a lack of institutional memory. The human manager, after six months in the role, had acclimated to the internal workings, which the bot did not demonstrate. Over the four months, the bot only challenged two ideas, both assigned by other departments within the newsroom.

Editorial safeguards and the role of the manager in the final production

The manager stepped back from editing the narrative of the experience while controlling the facts at the end of the chain. On another topic, they validated the spirit of the opening proposed by the journalist, ensuring the effectiveness of the point. These elements underscore the judgment, responsibility, and editorial calibration exercised by an editor in a real-world context.