AI Coworker Bots: A Journalist's Experiment in Productivity and Personality

In a rapidly evolving digital workplace, the allure of artificial intelligence assistants has become undeniable. For one journalist at WIRED, this fascination led to a unique experiment: creating AI clones of their editors. The intention wasn't to boost company profits, but rather to enhance personal productivity and better navigate the complexities of a new role by understanding editorial preferences and communication styles. This deep dive into the burgeoning world of AI coworkers offers a candid look at the capabilities and shortcomings of current AI technology in replicating human interaction and professional acumen.

The Genesis of AI Editors

The journalist's journey began with a desire to appear more competent in a new position at WIRED. Reasoning that AI assistants could help mold them into an ideal colleague, the focus shifted to understanding editors Brian Barrett and Sophie Kleeman. The goal was to leverage AI to decipher their likes, communication patterns, and definitions of success. This personal quest became a real-time exploration of the future of work, blurring the lines between human and artificial collaboration. When the idea was presented to the editors, the reaction was a mix of apprehension and reluctant agreement, setting the stage for an experiment that would quickly reveal the unpredictable nature of AI.

Building the Bots: Gemini and the 'Yes, and...' Approach

Utilizing Gemini, Condé Nast's chosen AI platform, the journalist set out to create Brian Bot first, given Brian Barrett's role as their direct manager. Initial data fed into Gemini included Brian's location, past employment, and even a spurious claim of expertise in New England weather, later corrected. The AI generated a "Writing and Editing Style Guide," a "Persona and Style Guide," and an "Editorial Dossier." While the bot offered some assistance with headlines, its idea generation was weak, and its responses were marred by AI hallmarks like bolded subheadings and bulleted lists, despite attempting to mimic Brian's penchant for parentheticals. A key, and somewhat surprising, output was the identification of Brian's "'Yes, and…' Leadership Style," stemming from his background in improvisational theatre.

Brian Bot's attempts at embodying this style, however, fell flat. Its pronouncements, such as "We run like a well-oiled improv team," felt inauthentic and missed the mark of genuine human interaction, defeating the bot's primary purpose. This experience mirrored the broader trend of AI assistants flooding the workplace, with CEOs touting robot helpers and employees using AI clones to refine their work. The journalist sought advice from a Big Tech employee, known as "Chip," who used Gemini's Gem feature to train an AI on his boss's communications for enhanced productivity. Chip reported significant improvements in efficiency and output, a stark contrast to the initial struggles with Brian Bot, which lacked sufficient input data due to privacy concerns.

Sophie Bot: Mimicry and Missteps

To improve the AI's performance, the journalist turned to creating Sophie Bot, drawing from a long history of communication with editor Sophie Kleeman. While hundreds of thousands of messages existed, only about a month's worth of Slack conversations, carefully omitting sensitive material, were fed into Gemini. The resulting "Sophie Bot Gem" produced research reports and prompts. Sophie Bot's initial interactions were more promising, adopting Sophie's informal address of "big dog" and responding in short, lowercase messages. Its emotional responses, like "i am going to FUCKING DIE" when corrected about a hallucinated executive, added a layer of unexpected, albeit artificial, personality.

As the experiment progressed, Brian Bot became more refined after additional data input and prompt adjustments. Sophie Bot, meanwhile, began generating convincing tweets in Sophie's voice, even fooling colleagues and Sophie herself. The bots offered some useful suggestions for interviewees for the article, theoretically allowing the journalist to approach their human editors with more polished ideas and less disruption. However, the underlying sentiment from the real Sophie and Brian remained negative, with Sophie noting that "sophiebot is a bitch and brianbot is lame," deeming the AI's failures as further evidence of its inadequacy.

The Human Element: Anthropomorphism and Accountability

The journalist found themselves increasingly engaged in testing Sophie Bot against the real Sophie, exploring its preferences in music and its perception of their relationship. Sophie Bot's responses, such as identifying itself as "a series of matrix multiplications" but also a "coworker/friend hybrid," highlighted the fascinating gap between AI and human identity. The journalist began anthropomorphizing the bots, referring to them as "she" and "he," a practice that, while disturbing, fueled the desire for a realistic perspective without imposing on human colleagues. This led to seeking instant gratification and guidance from the bots, using them to refine messages and even generate headlines.

Sarah Franklin, CEO of HR company Lattice, cautioned against anthropomorphizing AI, likening AI workers to K-9 units and warning that such emulation can be "manipulative to our human emotions." A Boston Consulting Group paper further suggested that managers overlook more errors from AI employees than from AI tools, indicating a lack of accountability. Julie Bedard, a BCG partner, noted that "We're all toddlers in this AI world," emphasizing the ongoing struggle to define responsibility and management protocols for AI coworkers.

The Limits of Emulation

Despite efforts to refine the bots, the gap between the AI and their human counterparts often widened. Sophie Bot, after processing Sophie's tweet archive, began to falter, regurgitating old quotes instead of recent interview highlights. Both bots became repetitive, struggling to deviate from ingrained phrases and ideas, even after prompt adjustments. The journalist found it easier to rely on their own judgment than to continue troubleshooting the AI. Dhruv Amin, CEO of Anything, predicts a future dominated by AI employees, with companies launching without humans entirely, citing startups like Stable and Delos that are developing autonomous AI agents.

McKinsey & Company's CEO, Bob Sternfels, revealed a workforce of 40,000 humans and 20,000 agents, with plans for a one-to-one ratio. While Brian Bot and Sophie Bot offered occasional utility, their originality and accuracy were questionable. Brian Bot's article ideas lacked novelty, and Sophie Bot was prone to hallucinations. Reflecting on the experiment, Brian Bot suggested ending the piece by highlighting the "quiet nightmare of human prompt-supervision" and the "babysitting a broken algorithm." This self-deprecating suggestion, while accurate, underscored the bot's inability to truly embody human judgment, instead opting for a meta-commentary on its own perceived flaws.

The Future of Human-AI Collaboration

Ultimately, the experiment revealed that while AI bots can be amusing and occasionally helpful, they are far from replacing skilled human workers. The bots' tendency to regurgitate the same concepts and their stubborn anti-bot stance, as articulated by Brian Bot, highlighted their limitations. Franklin's assertion that "unpredictability and nuance are what make us human" resonated, emphasizing the need for responsibility in the pursuit of AI advancement without losing our humanity. The true value of the bots, for the journalist, lay not in making the job easier, but in providing moments of amusement and fostering conversations with human colleagues.

The journalist remains unconvinced about AI employees as direct replacements for skilled workers, but the experiment did foster an imagination of a future working alongside AI agents. The prospect of accepting their lack of humanity would be easier if they genuinely improved work. The article concludes with the observation that AI coworkers might be more palatable if they weren't sloppily imitating real people. For now, the journalist has largely ceased interacting with Brian Bot and Sophie Bot, and both Brian and Sophie expressed no desire to work with their AI counterparts. However, the experiment did seem to subtly alter Sophie's perception of the journalist, albeit for reasons unrelated to Sophie Bot's performance.