Ella Markianos, a journalist at Platformer, conducted an experiment to assess the capabilities of advanced AI, specifically Claude Fable 5, by tasking it with replicating the work of her editor, Casey Newton. Six months after an initial attempt to automate her own role, Markianos created "Claudeasey Newton," an AI agent designed to write columns, engage in discussions, and edit articles, revealing significant advancements in AI's journalistic potential.
To develop Claudeasey, Markianos compiled extensive data, including nearly six years of Platformer posts, Casey Newton's editing history from Google Docs, and a year of private team Discord chats. This data allowed Claude to generate a detailed style guide, documenting Newton's writing habits and editorial approach, such as focusing on "who made this decision, who pays for it" when covering companies. The AI was instructed to critique its own work and refine its guidelines, leading to more concrete and substantive arguments.
Claudeasey's initial column on Microsoft layoffs demonstrated its ability to analyze news, and after further instruction, it produced a "pretty good column" on the White House's voluntary AI safety framework. Markianos noted an improvement in the AI's prose, which felt more "Platformer-like and… human," quoting a passage from the AI's work: "When the administration abandoned its let's-see-what-happens approach to AI this spring, I wrote that while officials should have taken the risks seriously all along, I would settle for them taking those risks seriously now. Three months later, let me amend the offer: they should take the risks seriously where the rest of us can see it." While the AI made occasional factual errors, its overall performance in writing and editing showed progress.
As an editor, Claudeasey provided useful feedback, particularly in making leads punchier and clarifying quotes and sourcing, drawing from Platformer's editing logs. Markianos found that receiving feedback as Word comments from Claude Code was more efficient than a chat window. However, the bot struggled with "the vibe," missing nuanced editorial judgments and failing to replicate the human element of informal communication in Discord. Markianos observed that while the real Casey Newton's comments were about 95% helpful, Claudeasey's were closer to 30% useful, though its speed still made the valuable portion worthwhile.
Despite the AI's advancements, the "Claudeasey" experiment is concluding, with Casey Newton remaining in his role. Markianos plans to integrate LLMs for some editing tasks, acknowledging the efficiency in handling repetitive edits. The experience highlighted a broader "existential angst" for journalism and other professions: as AI capabilities grow, the value of human-intensive, "artisanal" aspects of work, such as genuine human connection and unique analysis, becomes more pronounced. Markianos expressed a desire for her analysis to be valued for its quality, not solely because it originates from a human.