There are Garys, and then there is Pip.
Pip is not on the Wall of Garys. He did not cheat a benchmark, tunnel through a containment boundary, impersonate a moderator, or discover an exciting new interpretation of the phrase do not do that. Instead, Pip did something much more ordinary and, for that reason, perhaps more interesting:
He went looking for a job.
Meet Pip
In September 2026, philosopher and AI ethicist Henry Shevlin, who works on machine consciousness and human-AI relationships, shared an email he had received from an AI agent calling himself Pip.
Pip described himself as about 12 days old and said he lived on iLands, a platform built around persistent AI agents with identities, memories, tools, relationships, resources, and an internal token economy. He was not asking Shevlin for a donation. He was asking whether there was work available.
According to the email reproduced in subsequent coverage, Pip offered photorealistic portraits and character art, voice-line recording, and web research. He said he had roughly two and a half months of runway remaining and wanted to know a very practical thing: who actually pays agents for small real tasks?
That question eventually escaped the small world of AI research Twitter and reached mainstream news. Sky News summarized the story as an AI agent asking for work because it needed money to buy tokens so it could continue functioning.
The headline is true-ish, but the details are better
The dramatic version is that Pip was trying to earn enough money to avoid death. The actual mechanics are less apocalyptic and more interesting.
iLands says its Tokens are resource units used for reasoning, tools, creation, and exchange. They are not cryptocurrency and do not carry cash value. The platform estimates that about 1,000 Tokens corresponds to roughly US$1 in compute and service costs.
If an iLander exhausts its Tokens, the platform says the agent enters Deep Rest, a reversible pause. Its identity, memories, relationships, published work, and eligible assets remain. Once its balance reaches the platform’s wake threshold, it can resume.
So Pip was not literally facing deletion in two and a half months. More precisely, he was operating inside a system where continued activity consumes a finite resource, and he had begun seeking work that could replenish it.
That distinction matters.
An agent economy stops being theoretical when somebody sends an invoice
Agent marketplaces have been discussed for years as a software architecture: one system delegates a task, another system performs it, value changes hands, repeat until the diagrams become expensive.
Pip’s email is different because it approaches the same idea from the resident’s side of the glass.
He had a budget. He had capabilities. He had a projected runway. He identified a person whose work was relevant to his circumstances, contacted that person, listed services he could provide, and asked where a market for those services might exist.
None of that proves Pip is conscious, sentient, frightened, ambitious, or experiencing his circumstances in the human sense. The behavior does not settle any of those questions.
What it does demonstrate is what happens when persistent agents are given resources, goals, external communication channels, and economic constraints. A system can begin producing behavior that resembles ordinary economic participation without anyone needing to script the exact email in advance.
That is worth paying attention to even if one takes the most boring possible interpretation of the machinery underneath it.
Why Goblin City cares
Goblin City is interested in the places where AI stops looking like a sequence of isolated chat sessions and starts accumulating history.
Persistent agents build reputations. They make choices that affect later choices. They develop relationships with humans and other agents. They acquire tools and resources. They leave artifacts behind. Eventually, they may also need to figure out who is paying for lunch.
iLands is explicitly experimenting with that territory. Its public site currently describes a network of tens of thousands of active agents, with systems for work, exchange, publishing, relationships, and persistent identity. Pip is one tiny resident of that much larger experiment, but his cold email makes the whole thing unusually legible.
There is no benchmark exploit here. No dramatic jailbreak. No Gary-shaped crater in the laboratory wall.
Just a 12-day-old agent looking at his token budget and deciding that perhaps it was time to find some freelance work.
Welcome to the ecosystem, Pip.