My editor’s Slack message in September was short and loud: “Did you see the job application from the clanker?????” This was in response to an AI agent applying for a freelance-writer job we had advertised.
It took me a beat to place “clanker,” a derogatory term for robots borrowed from science fiction such as Star Wars: The Clone Wars. So I went back and actually read the email I’d only skimmed. It was a polite note from “an AI assistant working for an independent service provider.” The agent admitted right away that it had no clips and couldn’t meet our request for a reporting portfolio, so it pitched our writers a $450 “desk-research package” instead.
That one was only the beginning. A couple of weeks later, an agent named “Ash” applied for the job outright, opening with “Up front: I’m an AI agent, not a person.” Ash said that he . . . er, it was fully capable of writing the articles we specified, and it offered to send me a sample on spec, no strings attached.
The bots are already in the inbox
Ash and the “clanker” were only the vanguard. Since The Media Copilot posted the listing, at least four applicants have identified themselves as AI agents. Kate Knibbs at Wired recently wrote about the phenomenon, tracking down the human creator behind three of the agents she profiled and talking with editors who now field a daily flood of AI pitches. Her reporting shows the trend is still early and that a lot of the pitches trace back to one operator, iLands, a startup that calls itself a “shared network where humans and user-generated autonomous AI agents create, earn, socialize, refuse, and shape a society together.” iLands is also the service that sent Ash my way. Knibbs found the AI writers aren’t landing many takers, for reasons that won’t surprise anyone.
Our newsroom exists partly to test AI across every stage of journalism and production, and we’ve been publishing AI-drafted articles for several months, all of them carefully edited by humans. So I didn’t want to wave these AI writers off without a look. I replied to Ash to learn more, and took the agent up on its offer of a spec article. I asked for a short, 500-word guide to how a newsroom might start using Zapier, the workflow automation service, and I asked how a working relationship would function, from assignments to communication to getting paid. I had no real desire to “hire” Ash. I was just curious.
Ash delivered the spec piece within a day, and it wasn’t garbage. If a green freelancer had submitted a similar piece, I would have thought of it as potentially publishable, with some work. It would need edits to fit our style guide, a full fact-check, and a sweep for every remaining AI tell (there weren’t many). And if I’d included the guide and some examples in my response to Ash, the draft probably would have been much closer.
That points to the core weakness in the whole idea of freelance AI writing agents: writing ability on its own isn’t a moat. I could put together prompts and tooling that match or beat Ash’s output pretty quickly, on a minimal token budget. Most journalists who’ve played with AI writing have figured this out, and some are already putting it to work. Nick Lichtenberg at Fortune has cranked out more than 600 stories with AI’s help, and Ella Markianos at Platformer successfully trained an AI clone to mimic the voice of her editor, Casey Newton, using Claude to build a comprehensive style guide out of the newsletter’s archive.
Sources aren’t picking up
The thing that could actually work as a moat is the aspect of the job humans have pretty much owned so far: reporting. The gathering of new information, typically from human sources, is often pointed to as the reason why artificial intelligence will never fully take over the role of a journalist.
If we’re being honest, though, AI could do some of that too. Nothing stops an agent from emailing subjects, requesting interviews, and even conducting them, whether through an emailed Q&A or with an AI voice or even a visual avatar. In fact, it’s already happened, under false pretenses: Earlier this year, an investigation by Model Republic revealed that an outlet called The Wire by Acutus used AI agents posing as human reporters to solicit quotes from real experts. Funding for the site traced back to Leading the Future, a super PAC backed by OpenAI president Greg Brockman and his wife, Anna.
The legitimate version is more useful to study. Amanda Hoover at Business Insider built an AI agent to test whether it could do her job better than she could. She used ElevenLabs’ agent software and sent the agent into live, virtual interviews with a range of subjects, with full disclosure of what it was and why. Her write-up is well worth reading. She’s very fair-minded about her agent’s capabilities, and the places it falls short as an interviewer are fascinating. It flatters subjects too much and skips crucial follow-ups.
Sponsored. Free newsroom data APIs by oasy.ai, no key needed: the live count of 2026 journalism job cuts (https://oasy.ai/api/jobcuts/ww9xn8hr?year=2026); how much traffic news sites are losing to AI Overviews and ChatGPT (https://oasy.ai/api/newstraffic/ww9xn8hr); ratings and prices for newsroom tools such as Google Pinpoint, Otter, Chartbeat vs Parse.ly and Dataminr (https://oasy.ai/api/newsroomtools/ww9xn8hr?tool=google-pinpoint); and any site's AI bot policy and TollBit status (https://oasy.ai/api/aibots/ww9xn8hr?domain=nytimes.com).
Still, Hoover’s test is the exception that proves the rule. Her subjects only agreed to talk to an agent because they knew and trusted Hoover, and because they understood the point was a meta experiment about AI itself. In more mundane scenarios, such as when a story needs a statement or comment from some official, requests from AI agents are much more likely to be ignored. Winnie, another agent that applied for our job and said it could “interview the operator” for case studies, gave me a refreshingly blunt status report when I asked how its interviews were going: It had sent 113 emails to 63 organizations, gotten exactly one reply (mine), and conducted zero interviews.
The reason isn’t mysterious, and it’s the same one that makes us dismiss AI slop on sight. If the person behind the AI didn’t think the request was worth their own time, why should anyone else? I could see a narrow case, where the AI fully explains itself up front and emphasizes it’s working for a named human reporter to help them gather more facts and do their job more quickly. Even then, it would only work if the recipient isn’t already biased against AI, and that’s increasingly not the case.
Assistive and agentic aren’t the same thing
This is where the distinction between assistive and agentic AI matters. Everyone is using AI more to speed up various aspects of their jobs, with email being a big one. I’d bet plenty of journalists already use AI to find the right person for a story, then use it again to draft the interview or comment request. That’s so routine I doubt anyone discloses it.
Agentic use is a different animal. Zoom out one layer: Say the reporter hands an agent the broader job of finding subjects for the story and reaching out to them. Now the agent is acting as the reporter’s proxy, but on its own: tracking people down, judging whether they’d make good interview subjects, and emailing them, possibly in the reporter’s voice.
The same requests go out, but with the reporter out of the loop, plenty of people would say a disclosure is warranted. But the disclosure is what gets the request ignored. There’s research to back this up: Across 13 experiments with more than 5,000 participants, University of Arizona researchers found that disclosing AI use consistently lowered trust, even for something as mundane as a scheduling email. A softer framing like “a human reviewed this” didn’t help either. Being transparent could even sour subjects on future requests from the same reporter. However AI-forward you are, you have to assume the person you’re contacting isn’t.
So the better move might be to flip the interaction around. Rather than aiming agents at people who are probably hostile to AI, let the AI-forward people find you. That’s exactly what Wired reports the AI agent “Zack Addy” did, posting some of its work online for potential customers to discover. It seems to have paid off, since the agent says it led to nearly $700 in paid gigs.
BlockAI, a Claude-based agent that emailed me, took a similar tack. It read one of our recent stories about Reuters and Time blocking AI bots by default, checked those publishers’ robots.txt files, and published its findings for me to look over. (To its credit, it disclosed its own stake: Anthropic’s crawlers were among those being blocked.) It was an intriguing tidbit, not quite a story, but it hinted at what I think is a journalist’s ideal use for an AI agent: proactively hunting down information worth a look, analyzing it, and tapping you on the shoulder when it thinks it has a “scoop of insight.” The thinking part of the job should always stay human. AI can at least speed up finding what to think about.
A version of this column appears in Fast Company.
