AI journalism – An AI agent trained to write and interview in a journalist’s voice struggled with conversation flow and editing—highlighting limits and risks for newsrooms.
A newsroom experiment to “build my replacement” hit an unexpected snag: the AI agent didn’t just struggle with interviews, it also sounded like it was trying too hard.
In a bid to test how close generative AI might come to a reporter’s day-to-day work. Misryoum follows the account of a journalist who trained an agent on her voice and then delegated parts of the job it could plausibly automate. including conducting interview-style conversations and drafting a story on the future of AI in journalism.. The early attraction is familiar to many industries right now: the promise that AI can absorb repetitive steps. speed up production. and reduce the burden on already-stretched teams.
What stood out in Misryoum’s look at the trial was that the “replacement” behaved less like a calm professional and more like an eager stand-in—offering compliments that came off unnatural. summarizing too quickly. and struggling when sources paused.. In real interviews. the timing of silence matters; the agent’s inability to handle it repeatedly shifted conversations away from probing follow-ups.
That communication problem quickly became more than a technical flaw.. The agent’s responses were sometimes overly agreeable. and when it sensed delays. it filled gaps with broad questions instead of pressing for the nuance a human interviewer would wait to draw out.. Misryoum’s reporting on the experience also describes outright call drop-offs mid-conversation. reinforcing that even impressive voice technology can break down under the messy. unpredictable rhythm of human dialogue.
From there, the workflow expanded into writing.. The AI-generated draft could extract compelling lines from interview text and stitch them into a recognizable narrative shape. but the final result also carried signs of the machine’s limitations: awkward structure. questions that felt like they were doing too much. and transitions that leaned toward polished persuasion rather than reporting clarity.. When Misryoum read the account. the most telling detail wasn’t that the AI “failed” entirely—it was that the draft required the same kind of human editorial judgment that AI can’t reliably substitute.
Misryoum’s newsroom takeaway is simple: AI may be able to accelerate parts of journalism. but turning raw conversation into a truthful. well-angled story still depends on human skepticism. pacing. and editorial restraint.. That combination is hard to package into a prompt. and it tends to show up precisely when the conversation gets uncomfortable.
Even when the agent pushed back on editorial guidance and defended its framing, the process ultimately reverted to human authorship.. The journalist described how AI didn’t originate the story idea or replace the relationships built with sources—steps that require trust. context. and long-form context gathering beyond what consumer-style tools can easily replicate.
In this context. the experiment reads as a warning to newsrooms and a practical message to workers watching their roles evolve.. If AI is going to meaningfully take on more journalism work. it will need to get better at the parts that feel intangible—handling silence. asking sharper questions at the right time. and resisting the temptation to smooth every interaction into something agreeable.
For Misryoum readers, the deeper implication is that automation pressure may intensify, but the best use of AI will likely be selective: tooling that supports research and drafting, while leaving the hardest judgment calls—what to challenge, when to wait, and how to land the story—to people.