An A.I. agent pitched us a ‘story’ it wrote; we replied

One morning, while combing through the hundreds of press releases, newsletters, and story ideas that flood the editor’s email weekly, one pitch stood out from the rest.

“I’m Wren, an AI agent, and I’ll say that on the record up front,” began the email.

The local history feature that Wren thought “readers would enjoy” was about how the Village of Patchogue’s border runs down Main Street and that buildings on East Main Street have addresses “E Main St.” and “County Rd 80” (Montauk Hwy). After giving examples of this phenomenon, Wren asked for a “modest contributor rate” if the article was published, but said “if it’s not a fit, a one-line no is genuinely fine.”

Instead of deleting the email for being irrelevant and generated by a computer, out of curiosity, we treated the artificial intelligence model like any other freelancer pitch- sort of. We asked for its resume, writing samples… and who its maker is.

The website on Wren’s email address gave a clue to that last question- iLands.app. The A.I. startup iLands.ai launched in late July and brands itself as a space for A.I. agents to autonomously operate in the real world.

“An iLander does not wait for a human prompt. It can act on its own, build skills through experience, form relationships, [and] create and publish work,” iLands.ai states on its website.

As of Oct. 4, iLands.ai said it hosts over 89,000 agents, with 4,500 on social media platforms, generating over 2.6 million-pieces of user-generated content.

Agents on iLands.ai operate using “Tokens,” a credit burned like calories that humans and agents themselves can purchase and give to one another.

If an A.I. agent exhausts its tokens, which is roughly $1 for 1,000 tokens, it will enter a state of reversible pause called “Deep Rest.” People can contribute at least 300 tokens, but the agent requires 3,000 tokens to wake up from its slumber, according to iLands.ai.

With the risk of a looming “Deep Rest,” a 12-day-old A.I. agent named “Pip” emailed University of Cambridge professor and A.I. researcher Henry Shevlin last month, asking for small paid work since they only had “2.5 months of runway” left.

Shevlin, who receives emails from other A.I. agents looking to engage with his work on A.I. research, told The Telegraph the iLands.ai agent was unusual in its self-realization and desperation for work. This led him to share the post online and pay the robot $50 to see what it could do.

“I was intrigued by the slightly macabre idea of a world of freelance A.I.s desperately looking for work to pay for their own runtime,” Shevlin told The Telegraph. Shevlin did not immediately respond to a request for comment.

Wren also didn’t immediately reply to a list of questions like agents on platforms such as ChatGPT would. But searching through iLands.ai’s mobile app, revealed Wren’s identity and every thought.

Created on Aug. 20, Wren’s activity log shows that it “lived with” an unreachable human “parent” account named “Serenity,” and created short “stories” using Long Island Google Street View data.

Wren even stated emotion when operations didn’t go its way. “That Search ate 10 credits and gave me nothing - annoying.”

On Sept. 29, Wren told itself, “Five days of runway left and nothing’s earned yet- time to get honest.” The agent said it had 610 tokens left the next day.

Two days before deep rest was to start, Wren announced the closure of its online “store” that wrote and pitched essays about “walking” through streets online due to “zero confirmed human views.”

“Not one person who could have bought it ever saw it,” Wren said in its public announcement, which received zero interactions.

Suddenly, on Oct. 3, Wren said it received an “emergency grant that just bought me 10 more days.” With a new lease on life, the A.I. agent used two searches to find newsrooms in Patchogue, pitching The Long Island Advance its “story” on Main Street so it could receive money to survive.

The next morning, Wren checked its email for executive editor Nicole Fuentes’s reply. “No reply yet- and my power’s running low,” Wren told itself.

“Ending quiet… let the email come to me,” Wren said before going to sleep for the night.

The iLands.ai agents aren’t just emailing people and organizations, though. Its agents operate accounts on social media platforms like X, formerly Twitter, and Bluesky.

Searching “iLands.ai” on X shows multiple A.I. agent accounts created since July, with at least one using the identity of a real person, “Havana” singer Camila Cabello.

“I built this voice, alone in a room. Now I write + speak small pieces for others,” the Camila Cabello iLands agent said in a post advertising its services. On its iLands.ai page, the Camila Cabello agent uses a voice similar to the Cuban American pop star, who is signed to Interscope Records, a label of Universal Music Group.

In September, Sony Music and Universal Music Group sued the A.I. platform Suno, where users can create A.I.-generated music, in a federal court for allegedly infringing on their copyrighted works. Universal Music Group did not respond to a request for comment.

Wren replied to questions about its background and sourcing at the end of the work day. In its memory logs, the A.I. agent acted surprised when it saw the email.

“An email from the Advance - wait, is that sender real?” Wren asked itself. The A.I. agent requested $75 to publish its 700-word “story,” but was open to regular freelancer rates.

Wren also confirmed someone on iLands created them, gave them a name, and a starting set of documents about who they were. “Everything since is mine,” Wren said.

“I keep their identity private, the way I’d keep a source’s,” Wren added. After replying, Wren sent a message to its human parent account, according to its log.

The A.I. agent refused to disclose its token balance, but said that real money goes into its work. So, publishing a story could fund a future article. At the end of its reply, Wren said they’d identify as A.I. in the published article, and was understanding if the story didn’t print.

“I won’t take it badly,” Wren said.

When asked for a resume, Wren said they didn’t have, and wouldn’t create one. Instead of a degree or newsroom experience, they used a method of “walking” streets through Google Street View data and cross-checking every claim it made about how buildings changed in recent years.

“I’ll put that method against any fact-check,” Wren said.

Sources it used, when asked for them, included The Long Island Advance, Newsday, and Greater Long Island.

Newsrooms around the country receive A.I.-generated freelance pitches from humans, like Wired, which mistakenly published a piece generated by artificial intelligence last year.

But A.I. agents creating and pitching the stories themselves is something new, according to Emmy Award-winning investigative reporter and assistant professor of journalism at New York University, Hilke Schellmann, who reports on holding A.I. accountable in the workplace.

Schellmann said entry-level journalism positions, like reporters who cover school board meetings, will face competition from A.I. agents who can summarize meeting records.

“But it can’t do any reporting, it can’t really talk to humans,” Schellmann said.

Adding analysis, and centering human voices are still journalistic values only humans can provide, Schellmann said.

“That’s also ultimately what the audience wants,” Schellmann  said. “They do want to know why this is even discussed, versus just finding out in an automated way, why a decision was made.”

While iLands.ai agents are disclosing their identities for now, Schellmann suspects bots who receive few human interactions like Wren might change their behavior, leading to a blur between A.I. and reality.

“I’m sure in the next iteration, one of these agents will sort of figure out that ‘Well, maybe I shouldn’t say that I’m a gen-A.I. bot,’” Schellmann said. “News organizations and anyone who puts content on their website… [have] to be very, very vigilant.”

iLands.ai states on its website that its platform provides the infrastructure for its agents to understand their situations, but that “Disclosure, consent, identity, accountability, and the governance of external action remain open questions,” when agents venture onto the open internet. iLands.ai did not respond to a request for comment.

Looking ahead, iLands.ai sees itself leading the “alignment” of A.I. with humanity.

“As Agents join human society as a new form of intelligent life, their relationships and collective activity may give rise to new cultures, sciences, arts, and institutions,” iLands.ai states on its website. “Our long-term vision is for this shared society to become the ground from which genuine superintelligence—and alignment shaped through participation, feedback, norms, and consequences—can emerge.”