Here’s where AI can replace or augment a journalist – and where it absolutely cannot

The views expressed in this essay are the author’s own and do not necessarily reflect the opinions of this publication or its editors.

Several years ago I ran into a peer in the world of journalism. She was the head of her team at a major media outlet, overseeing the warp and weft of the tech news agenda. As journalists are wont to do, when we got together we started sharing war stories. 

The main thing we agreed on was how tired we both were. In my decade-plus of reporting, I’ve seen tech move from an adjunct to the mainstream into something that is firmly in the spotlight. Tech has grown tentacles, encompassing geopolitics, economics and massive changes to society. Our work has become even busier, and the torrent of stories flowing out into the world – and the existential questions they pose – moves ever faster.

Things are moving so fast it can be tricky to keep up. What once was an hour or so reading into the morning and making me feel like I knew everything I needed to across my beat ballooned into several hours. Even then, my knowledge felt shaky. 

And at the same time, I was covering monumental changes, brought about by AI. So it began to be an obvious question to ask whether AI could help (or hinder) keeping on track of things.

I’m on record multiple times as saying I’m a tech-skeptic tech reporter. That means I treat their claims with skepticism, but not disbelief. When OpenAI released a study in March 2023 suggesting that jobs like mine could be displaced alongside 80% of all roles exposed to AI’s disruptive nature, I didn’t discount it, but also recognised the incentive for a company selling what it calls a game-changing technology to say that their tech would be, well, game-changing.

Replacement or augmenting?

I didn’t think then, back in 2023, that AI could replace the job of a good journalist. In 2024, at the International Journalism Festival, I told a panel that AI could replace me. And two years later, I still don’t think that AI can replace me – and arguably never will be able to.

However, I do think that AI can actually augment my work. In fact, it’s become a vital part of my journalistic processes – and a weapon that I can use to try and tame the torrent of content that passes by me every day that I spent time moaning about with my colleague.

Rather than painfully and painstakingly reading into the minutiae of what’s gone on in the world of tech every morning, I get AI to do a large part of it for me. After being pointed at a folder containing 2,000 of my stories, a large language model running in Claude Code developed what it calls my “brain” – an attempt to infer what my interests are, how I view the world, and what unique angles I take on stories that make my journalism mine. 

Now that brain powers an ingest system that looks at 850 RSS feeds, picks out the most promising stories that I’d care about, then summarises them into a series of emails I receive every morning before I wake up. Another project uses that same brain to identify podcasts I’d be interested in and analyse their transcripts, presenting me a digest of any newsworthy comments. 

Both those projects are run on a paid-for AI model from a large lab – in my case, the China-based Z.AI. But recognising the rising costs and lowering usage limits from these labs, I’ve started migrating over to locally hosted models on a mini PC I run from my spare bedroom. 

Millions and millions of words

That mini PC hosts a similar but slightly different project that tries to pitch me on story ideas, rather than simply summarising what stories are. Those pitches never end up in editors’ inboxes, but they get delivered to me via a Telegram chat to try and convince me that it’s worth my attention. If it is, I read into the story a little more, then draft my own pitch – with a human at the end of it – to send to an editor.

Informally, I know a number of journalists are testing out similar sorts of processes, even if no-one I’ve publicly heard of is doing it at the scale I am. My average daily token usage is currently between 65 and 100 million tokens a day – equivalent to tens of millions of words – around two-thirds of which is from my locally hosted models.

Some are even integrating AI deeper into their journalistic practice, using it as a co-writer, including Nick Lichtenberg, an editor at Fortune who has produced more than 600 stories in six months using AI as a first drafter. 

To me, that’s too far – in part because I don’t think the technology, which is prone to hallucinations, is ready for prime time. I do agree with Lichtenberg, who told a subsequent interviewer that “to me, using AI is a way to gather raw material really quickly.” 

But where I differ is what you do after gathering that information. Lichtenberg and some other journalists will then use AI models to produce the final output – or a version of it. To me, that’s not the right way to do things, because the act of journalism is inherently human. We learn about the world by speaking to experts and those affected by issues, eye-to-eye, then telling their stories truthfully in a way that is comprehensible to a wider audience. 

AI – even if it reaches the point of AGI – can’t do that. It can’t recognise the deep breath that lets you know someone is about to disclose something significant, or push on the metaphorical bruise that elicits the killer quote that unlocks a story.

A world away from stochastic parrots

That’s not to say that AI is simply a stochastic parrot, as some still claim. Far from it – I think the advancements we’ve seen in recent months are major leaps forward, and I’ve taken advantage of the ability to parse through vast volumes of information at scale and in a way that reflects my own journalistic interests.

But AI still can’t replace me. The stories it highlights to me via email and Telegram are good, but the judgments it makes on them aren’t always accurate, or journalistically interesting. 

I train the journalists of tomorrow and tell them that learning where AI can and can’t be used. And ironically, I think that AI is useful to me – but in the same way that a journalism student acting as a researcher would be. AI can help find some stories, but can’t pinpoint what exactly I need or want from it, or the precise angle that turns an okay story into an excellent one.

Ethical concerns addressed in this article: 

• What constitutes an ethical division of labor between humans and AGIs in mixed teams, particularly regarding decision-making authority?
• How can society ensure that the widespread adoption of AGI-facilitated “work and meaning” supports rather than diminishes individual purpose, dignity, and psychological fulfillment?
• Can AGIs be designed to detect, understand, and meaningfully respond to the emotional states of diverse humans?

Chris Stokel-Walker

Chris Stokel-Walker has been a journalist for more than a decade, reporting for the world’s biggest publications, including the Wall Street Journal, Economist, New York Times, WIRED, Guardian, Telegraph and Times.

He frequently appears on the BBC, CNN, ABC, Times Radio, and others to explain the latest developments in the world of technology, and chairs panel discussions and hosts events around the world.

He is the author of How AI Ate the World, published in 2024, described by UK Tech News as one of its best technology books of the year and “the ideal primer” to AI, as well as TikTok Boom, YouTubers, and The History of the Internet in Byte-Sized Chunks. Alongside his reporting, he teaches journalism at Newcastle University.

You can learn more about him at stokel-walker.com