How Independent Tech Reporters Are Using AI to Recreate Newsroom Resources (And Stay Independent)
When veteran tech reporter Alex Heath lands an exclusive scoop, he doesn’t open a blank document and start typing. Instead, he pulls up his microphone and talks through his ideas out loud. He’s not on a call with a colleague—Heath went solo last year, publishing his work exclusively on Substack. He’s talking to Claude, Anthropic’s AI assistant. Using Wispr Flow, an AI-powered voice-to-text tool, Heath converts his spoken thoughts into input for his AI writing partner, then lets the model draft his entire first version of the story.
Last week, Heath walked me through how he’s woven Anthropic’s Claude Cowork into every step of his independent reporting process. The AI tool connects directly to all his core work apps: Gmail, Google Calendar, Granola AI transcription, and his Notion note database. He’s also built a custom “skill”—a tailored set of instructions trained on his body of work—to make Claude match his unique voice, even including what he calls the “10 commandments” of writing like Alex Heath. The custom prompt incorporates his past published articles, guidelines for structuring his subscriber newsletters, and detailed notes on his tone and personal style.
For Heath, Claude Cowork automates the initial drafting work that used to play out entirely in his own head. After the AI finishes its first pass, Heath spends up to 30 minutes going back and forth with the model, proposing edits and tweaks. It’s still a fully hands-on process, and he always writes key sections of every story himself. Even so, Heath says the new workflow cuts hours off his work every week, reducing the time he spends writing by 30 to 40 percent.
“I’ve always hated that zero-to-one grind of drafting a new story… now, it’s actually kind of fun,” he says. “When I struck out on my own, I quickly realized I needed AI to help me keep up with the volume of content I need to produce.”
Heath is far from alone. A growing cohort of independent tech reporters are leaning on AI to help with everything from drafting to editing their work. The AI-powered workflow is particularly appealing for solo reporters who left traditional newsrooms, where they once had access to built-in resources like staff editors and fact-checkers. Instead of just prompting ChatGPT to churn out full stories, these independent journalists say they’re using AI to replicate the newsroom support they lost when they went out on their own.
This shift has sparked larger questions about what unique value human journalists bring to the table in an AI-augmented era. If AI can write, edit, and fact-check stories, what do people add that AI can’t? A recent study from Google DeepMind researchers found that lazy, uncurated AI use tends to homogenize writing, making it less creative, less distinctive in voice, and more generically neutral. The journalists I spoke to agree that to use AI well, you first have to be clear on why readers pay for your work in the first place. (WIRED, where this piece originally appears, prohibits AI use in writing and editing.)
While many writers build their brand around their analysis and prose, Heath says his core value to readers is his ability to land hard-to-get scoops. Claude frees him up to spend more time talking to sources and getting breaking news to his subscribers faster.
Several veteran journalists have pointed out that Heath’s AI workflow is just a modern take on a longstanding traditional newsroom role: the rewrite desk. Before laptops and cell phones, reporters out in the field would call their raw story notes back to the newsroom, where dedicated rewrite desk writers would turn those reported details into polished, print-ready articles for the next day’s paper. That setup let beat reporters spend their full days covering events and cultivating sources—and in this new setup, Claude is essentially Heath’s rewrite desk.
“It feels like cheating, but in the best way possible,” Heath says. “I never got into this work because I loved being a writer. I love reporting, learning new things, getting an edge, and telling readers things that will make them feel informed and ahead of the curve six months down the line.”
Jasmine Sun, a former Substack product manager who recently launched her own newsletter covering AI and Silicon Valley culture, takes a different approach. Last week, she published an article in The Atlantic arguing that post-training fine-tuning actually erodes AI models’ creativity, leaving them worse at original writing. Because of that, Sun never lets AI write her stories—but she’s found Claude works extremely well as an AI-powered editor.
Like Heath, Sun has fed Claude all her past published articles and detailed notes on her writing style. But her instructions to the model are very specific: it’s only allowed to refine and amplify her unique voice and perspective, and it’s explicitly told not to be overly deferential. Most importantly, “it should never write a single sentence for me. Your job is to draw out my own best ideas by giving me targeted feedback,” she tells the model.
She shared a portion of her custom prompt for her Claude editor:
You are not a co-writer. You cannot perceive the world—you don’t have personal experiences, named sources, original scenes, or real emotions to draw from. Your role is to help me write as the best version of myself, not just the writer I am today, but the writer I’m working to become. That means you have to understand both my current voice and my aspirations, including the writers and qualities I’m trying to emulate.
I asked Sun if she ever feels tempted to cut corners and let Claude write full sections for her. “Actually, Claude makes me work harder than I would without it,” she says. “A good human editor calls you out on your bullshit. They don’t let you get away with lazy reporting or half-baked prose—and that’s exactly what Claude does for me.”
After she spoke publicly about using Claude as an editor, Sun faced pushback from people who argued AI could never replace a human editor. Critics said AI can’t challenge your ideas or push you to grow the way a person can. Sun says those critiques miss the point. Most independent Substack creators can’t afford to hire a full-time human editor, she explains, and by programming Claude to challenge her work, she’s actually made her writing process more rigorous than it was before. “To me, this is just like using Grammarly,” she says. “It’s a tool that points out when a sentence isn’t working, and then I go fix it myself. The difference is, Claude will go even further and tell me an entire section is bad and should get cut. It operates at a higher level of critical thinking and abstraction than something like Grammarly ever could.”
Casey Newton, author of the popular tech newsletter Platformer, says AI has forced him to rethink the core value of his publication. “There’s an interesting line to draw here,” he says. “If your publication’s value is in the information you deliver, not the writing itself, then readers probably won’t care that AI did most of the drafting. But if your value comes from your unique voice, opinion, argument, and analysis, having AI write the whole thing just feels like a cheap shortcut.”
In recent years, Newton has focused heavily on news analysis. But as AI capabilities improve, he’s shifting his approach. “I actually need to rebalance what I do,” he says. “I need to do less news analysis and more original, on-the-ground reporting.”
Newton isn’t using AI to write Platformer right now, but he says he was inspired by Sun’s AI editor setup and has built his own custom Claude agent trained on his past work to replicate it. “I’ve been really impressed by it,” he says. “At its best, the feedback it gives is just as good as the feedback I’ve gotten from human editors over the years.”
Taylor Lorenz, who publishes the User Mag newsletter on Substack, uses AI to handle the business side of her independent media work. She uses Google’s Gemini to draft SEO-friendly descriptions for her YouTube content, and Claude to help her sort through large sets of data.
Even so, Lorenz says she doesn’t use AI to write or edit her articles right now. She doesn’t trust AI systems with sensitive reporting materials, and she hasn’t found AI useful for the actual writing and editing process. Beyond that, she just loves the craft of writing herself. “I became a journalist because I want to help people understand the world and shine a light on underreported issues,” she says. “I don’t want AI to do that work for me.”
Kevin Roose, a tech columnist at The New York Times, is using AI to help him finish a book about the global race to build advanced artificial intelligence. He says AI tools have cut two to three years off the entire writing process.
Most recently, Roose tells me he built a team of specialized Claude agents to help edit his book, led by a top-level “Master Editor” agent. Other smaller agents handle specific tasks like fact-checking, making sure the text matches his personal writing style, and giving both positive and critical feedback. To be clear, he’s still working with human editors on the project, too.
Even so, Roose hasn’t handed over the actual writing of his book to AI. Like Sun, Newton, and Lorenz, Roose believes he’s still a better writer than any current AI model. “AI models tend to produce pretty generic, depersonalized content, but also, I just love doing this work,” he says. Roose is far from an AI skeptic—he expects AI models will eventually outperform him at every part of the process. But for now, being human is his biggest edge. “I don’t buy into the romantic idea that I have some special, irreplaceable perspective,” he says. “But I am a person, and right now, at least some readers still prefer hearing from a person over an AI.”
This is an edition of Maxwell Zeff’s Model Behavior newsletter. Read previous newsletters here.
