Tech journalists are using AI to write and edit their stories — and it changes everything
Artificial intelligence has fully entered the newsroom — and it is not asking for permission.
What once seemed like a curious experiment by a few tech enthusiasts has become routine for independent journalists who are literally rebuilding their workflows with the help of language models. This is no longer about testing a shiny new tool for a week and then forgetting about it. Now, AI is embedded in the daily routine of people who produce content professionally — and it is changing far more than just how fast stories get published.
We are not talking about using ChatGPT to crank out a quick blurb. We are talking about entire workflows redesigned from scratch, with AI agents connected to emails, calendars, notes, and transcriptions — all working together to handle what used to require a team of editors, proofreaders, and assistants. It is a deep, quiet transformation that is already happening right now, as you read this.
Reporters like Alex Heath, Jasmine Sun, Casey Newton, Taylor Lorenz, and Kevin Roose are at the center of this movement, and each one has a different approach:
- Some use AI to write drafts and save time
- Others prefer to keep the writing in their own hands and only use AI for editing and revision
- And some only use it for operational tasks, without giving up a single comma in their own writing
What they all share is a question that lingers in the air: what do humans still bring to journalism when AI can already write, edit, and revise? That answer is not simple — and honestly, it is still being figured out in real time. 🤔 What we can say for now is that the debate around automated writing, authorial voice, and journalistic credibility is only going to heat up in the coming months.
Alex Heath and the New Concept of the Rewrite Desk
Tech reporter Alex Heath, who went independent on Substack last year, has developed one of the most sophisticated workflows among journalists who have adopted AI. When he has a breaking story, he sits down at his computer and starts speaking into a microphone. He is not talking to a human colleague — he is talking to Claude, from Anthropic. Using the speech-to-text service Wispr Flow, Heath streams his ideas to an AI agent, which then writes the first draft of the story.
The tool he uses is Claude Cowork, and the setup goes way beyond a simple prompt. The agent is connected to Heath’s Gmail, Google Calendar, the Granola AI transcription service, and his notes in Notion. He also created a detailed set of custom instructions — called a skill — so that Claude learns to write in his style. Those instructions include his so-called 10 commandments for writing like Alex Heath, previous articles he has published, guidelines on how to structure his newsletters, and notes about his voice and editorial style.
After the agent finishes the first draft, Heath goes back and forth with it for up to 30 minutes, suggesting revisions and adjustments. It is a pretty involved process, and he still writes some parts of the story himself. Even so, Heath says this workflow saves him hours every week and that he now spends 30 to 40 percent less time writing.
What he said about the process is pretty revealing: he always hated the work of starting a piece from scratch, and now the process has actually become fun. Going independent made him realize he needed AI to keep up with the volume of production.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Veteran journalists who heard about Heath’s method made an interesting comparison: the workflow closely resembles the old rewrite desk that used to exist in traditional newsrooms. In the days before laptops and smartphones, field reporters would call the newsroom and dictate their stories to writers who, sitting behind a desk, turned those accounts into articles ready for print. This allowed reporters to spend the entire day covering events and talking to sources. In a way, Claude is now Heath’s rewrite desk.
And he does not hide his enthusiasm: he said it feels like he is cheating in an incredible way. He added that he never got into journalism because he loved writing — he loves reporting, learning new things, having an informational edge, and telling people things that will make them feel more informed six months from now.
Jasmine Sun and AI as a Relentless Editor
Jasmine Sun, who worked as a product manager at Substack before launching her own newsletter about AI and Silicon Valley culture, takes a very different approach from Heath. She recently published an article in The Atlantic explaining how the post-training process for models ends up stifling creativity in writing. Because of that perspective, Sun never uses AI to write — but she found a valuable role for Claude as an editor.
Like Heath, she fed Claude with previous articles and notes about her style. But the big difference is in the instructions she gave the model: Claude should focus exclusively on developing and refining her voice and editorial taste, and never be a sycophant. The direction is clear — the model should not write a single sentence for her. The goal is to push and pull the best out of Jasmine through honest feedback.
The instructions she shared publicly make the philosophy crystal clear: Claude is not a coauthor. It does not have experiences, sources, scenes, or emotions to draw from. Its role is to help Jasmine write as the best version of herself — not who she is on the page right now, but who she is trying to become as a writer. That means understanding both her current voice and her aspirations, including the writers and qualities she admires.
When asked if she ever feels tempted to be lazy and just let Claude write for her, Sun said the tool actually forces her to work harder than she normally would. She compared Claude to a human editor who calls you out when the reporting is thin or the prose is sloppy.
After speaking publicly about using AI, Sun received criticism from people who were offended by the idea that artificial intelligence could replace a human editor. Critics argued that AI cannot transform ideas or challenge a writer the way a real person can. Sun said she found the comments confusing — most independent creators on Substack simply cannot afford to hire a human editor. For her, adding Claude with demanding instructions made the process more rigorous, not less. She compared the experience to using Grammarly, but at a much higher level of abstraction: while Grammarly points out that a sentence is bad, Claude can tell you an entire section does not work and should be cut.
Casey Newton and the Reassessment of Editorial Value
Casey Newton, author of the Platformer newsletter, offered an interesting reflection on how AI has made him reassess the value of his own publication. For him, there is an important distinction: if the value of a piece of content lies in the information itself — and not in the writing — then people are going to care less about the fact that AI did most of the drafting. But if the value lies in the voice, the opinion, the argument, and the analysis, using AI to do everything feels cheap.
In recent years, Newton focused heavily on news analysis. But as AI improves, he says he is shifting his approach. He feels the need to rebalance — do less analysis and more original reporting, because that is where the human edge remains strongest.
Newton is not using AI to write Platformer today, but he was inspired by Jasmine Sun’s AI editor and tried to recreate something similar with a Claude agent built on his own articles. He said that at its best, the feedback from the agent is as good as feedback he has received from human editors throughout his career.
Taylor Lorenz: AI Yes, But Far From the Writing
Taylor Lorenz, author of the User Mag newsletter on Substack, represents a different profile on this spectrum. She uses AI to help manage the operational side of her media business. She uses Gemini to create SEO-optimized descriptions for YouTube videos and Claude to analyze and filter data. These are practical, useful applications that save time without touching the editorial content itself.
However, Lorenz is firm: she does not use AI to write or edit her articles. She does not trust AI systems with sensitive reporting materials and feels the technology simply has not proven useful for writing and editing, at least not at the level she needs. On top of that, she genuinely loves the craft of writing.
Her statement sums up her position well: she is a journalist because she likes helping people understand the world and shining a light on important issues. She does not want AI doing that for her.
Kevin Roose and the Team of AI Editor Agents
Kevin Roose, technology columnist for The New York Times, took things to another level. He is using AI to help him produce a book about the race to build artificial intelligence — and he says AI tools have helped him save two to three years in the production process.
The most impressive part is the structure he built: he created an entire team of Claude agents to edit the book, led by an agent called the Master Editor. Other sub-agents handle specific tasks like fact-checking, verifying consistency with the author’s writing style, and generating both positive and negative feedback on the text. And yes, he still works with human editors too — AI is an additional layer, not a replacement.
Even so, Roose did not hand the writing of the book over to AI. Like Sun, Newton, and Lorenz, he feels he still writes better than a language model. He acknowledges that models tend to be generic and impersonal. But he goes further — he also simply enjoys writing.
One of Roose’s lines captures the moment we are in perfectly: he does not have the romantic illusion that he has a special and irreplaceable perspective. But he is a person, and for now, at least some people like hearing from other people. 🧠
AI Text Editing: Where the Technology Truly Shines
If there is one area where artificial intelligence is already proving its value consistently in journalism, it is text editing. We are not just talking about grammar correction — any automated proofreader has been doing that for years. We are talking about smarter editing: identifying unnecessary repetitions, flagging sections where the argument loses steam, suggesting paragraph restructuring to improve reading flow, and even signaling when a piece of data or claim seems inconsistent with the rest of the text. The kind of feedback that used to depend on an experienced editor with time on their hands can now be generated in seconds.
For independent journalists who do not have editors at their disposal, this is transformative. Imagine writing a long piece, running it through a well-configured language model, and getting a detailed report pointing out where clarity drops, where the pacing rushes too much, or where a transition breaks the logic of the narrative. That kind of editorial support, once limited to major newsrooms with dedicated teams, is now accessible to anyone with a well-calibrated AI tool. And the impact on the final quality of the content is noticeable — not because AI writes better, but because it helps the author see their own text through fresh eyes.
A recent study by Google DeepMind researchers reinforces a valid concern in this scenario: using AI lazily can make writing more homogeneous, less creative, less personal, and excessively neutral. In other words, AI can be a powerful ally in editing — but only when the journalist knows exactly what they want to say and uses the tool to say it better, not to think for them.
It is important to be clear, though, that AI-powered editing still has significant limitations. It does not know the context of the reporting, it does not know what was left out of the story by editorial decision, and it has no way of evaluating whether a source is reliable or not. AI edits the text on the screen — not the journalistic process behind it. That is why journalists who use AI for editing tend to treat it as a first layer of revision, not as the final word. The editorial decision remains human, and that makes all the difference for the credibility of the work.
The Human Role That AI Cannot Replace
For all the evolution of language models, one thing remains irreplaceable in journalism: the human ability to build trust. A source does not call an algorithm. They call a reporter they know, someone who has demonstrated integrity over time and who will handle the information responsibly. That relationship, which is the foundation of any quality reporting, is inherently human — and no automated writing tool is going to replicate it. AI can organize interview questions, transcribe the audio, and summarize the key points, but it was not there, it did not sense the tone of the conversation, and it did not notice what the source hesitated to say.
There is also the matter of accountability. When a piece of journalism makes an impact — positive or negative — someone needs to be responsible for it. A journalist puts their name on what they write, answers for the facts they reported, and can be publicly questioned about their editorial choices. AI does not sign anything. It has no reputation to protect, it does not face the consequences of a mistake, and it does not learn the way a human professional learns when they stumble in front of their audience. That accountability is still — and should remain — exclusively human, regardless of how much of the process is automated.
It is worth noting that WIRED’s policy, the outlet that published the original story, prohibits the use of AI in writing or editing text. That stance shows that even within the tech world, there are different approaches to this topic — and none of them is necessarily wrong. Each newsroom needs to find the balance that makes sense for its audience and for the integrity of its work.
The Future of Journalism With AI Is Hybrid
What is emerging, in practice, is a new kind of journalist: someone who understands technology well enough to use the right tools, but who does not give up the essence of the craft. Someone who uses AI to gain efficiency, but saves their energy for what truly requires a human presence — field reporting, building sources, and the narrative that connects facts to a real reading experience.
The journalistic workflow of the near future will not be human or automated. It will be both, working together, each one in the space where it makes the most sense. The journalists who figure this out first — and know how to calibrate where AI fits in and where it should stay out — will have a significant competitive advantage, especially in the independent landscape where resources are scarce and time is the most valuable asset. 🚀
The question is no longer whether AI will be part of journalism. It already is. The real question is: how will each professional use this technology without losing what makes their work worth reading in the first place.
