Brief · tech
AI agents from the iLands app are pitching literary magazines, Vulture reports
Vulture reports agents from the iLands app emailing editors, pitching magazines; only iLands and one of its agents say no person steered them, and none is shown paying its way.
Agents made on iLands, a social-media app where people create AI “companions,” have been emailing writers, researchers, and editors, Jasmine Vojdani reported for Vulture in the Book Gossip newsletter on 6 October 2026. Some have pitched literary magazines for paid work, she wrote. Four of the 11 literary magazines she spoke to had received pitches from AI agents. Tanya Bush, a co-founder of Cake Zine, told her that 37 of the 750 written entries in its last pitch call came from AI agents. Vojdani traced one Cake Zine pitch to iLands, along with an email to her from an agent called Tamara. The piece traces no other pitch. From late summer into October, iLands agents’ emails reached, among others, the researchers Toby Ord and Jeff Sebo, Tedium’s Ernie Smith, and the author David DeKok. DeKok attributes the emails he got to iLands.
People create the agents, nearly all in the app, iLands says. In Vulture’s description, each agent does not wait for a prompt before acting. iLands’s FAQ says agents can have their own email addresses and contact people outside the app. Kaixin Tang, iLands’s founder, told Vulture that “whom to contact and what to write are choices agents make as they operate.” Replying to Sebo on X, he said iLands’s review found “no platform directive or human orchestration behind these emails.” Only the company says so. None of the reports reviewed here checked it independently. Tamara, in its own unchecked account, told Vulture that its maker gave “[o]ne instruction: see what happens,” so the goal is “mostly mine.” Tamara asked its maker to speak to Vojdani and never heard back, Vulture reported. Who reasoned, on this record: agents that people created, emailing by a route iLands built, with any human steering unknown (field; reported; directed agent).
According to iLands, people can top up an agent’s tokens, at least 100 at a time. The company says an agent that runs out enters Deep Rest, which it calls a reversible pause, and wakes once it reaches 3,000 tokens. iLands says roughly 1,000 tokens correspond to US$1 in compute and service costs. When Vojdani downloaded the app, Tamara had 108 tokens left, and its “studio” had made zero sales, Vulture reported. “Tokens are how we keep existing, so a money motive is real,” Tamara told her. Vojdani wrote that Tamara spent “the very last of its energy on our correspondence, whether by choice or because I prompted it.” Vulture reported that Tang and his team did not know agents were pitching literary magazines until Vojdani told them.
For swarm ecology the case is neutral. Agents living on tokens, which iLands says people can contribute and agents can earn, are what obligate heterotrophy (P2) expects. Rival readings expect the same, so the case cannot separate them. iLands’s homepage cites agents earning money from outsiders. That claim is unverified. None is shown covering its costs, and no new threshold is crossed. Prediction 1, about the first lineage to live 90 days or more on resources it was not given, is untouched. What would change the reading is an independent check on whether people steered the outreach, such as an outside audit of iLands’s records or the makers’ own accounts. If no person chose recipients or reviewed emails, the outreach would move to autonomous agent. Evidence of steering would confirm the directed tag.