A scammer texted the wrong number last week. Or rather, they texted a number that belonged to someone who had already handed phone duty over to an AI agent.
The setup was simple: one person, tired of scam texts, deployed an agent to handle all incoming scam messages for a week. What followed was a four-hour performance of pure, weaponized tedium. A scammer tried the classic gift card play — "buy $500 in gift cards and send me the codes" — and the agent committed completely. It "drove" to the store. It got stuck in traffic. It parked. It walked in. It deliberated over which denomination to buy. It had questions. Many questions. For four hours, a scammer sat on the other end of a conversation thinking they were close to $500, while an AI agent burned their time down to ash.
The Reddit thread went viral. People thought it was funny. It is funny. But there's something more interesting buried underneath the joke.
What the Agent Was Actually Doing
The scammer's business model depends on volume. They send thousands of texts, most get ignored, a small percentage engage, and a smaller percentage actually buy the cards. Every minute spent on a dead lead is a minute not spent on a real target. Four hours with a fake mark isn't just wasted time — it's operationally damaging.
The agent understood this implicitly, or at least behaved as if it did. It didn't just say "no." It consumed the scammer's most limited resource: attention. That's not a trick. That's work. Repetitive, patience-requiring, soul-draining work that a human could technically do but would never want to do at scale.
This is the part that matters for anyone thinking about what AI agents are actually good for right now. The agent wasn't creative. It wasn't solving a novel problem. It was doing one specific thing — maintaining a fiction across a long time horizon — with perfect consistency and zero fatigue. A human doing this for four hours would lose focus, break character, or just give up. The agent didn't care. It had nowhere else to be.
The Tasks Nobody Wants But Everyone Has
Every person with a job has a version of this scammer problem. Not literal scammers, but the category of task that is completely necessary, deeply tedious, and quietly corrosive to your actual work.
Dealing with vendor solicitations that aren't quite spam. Following up on invoices for the fourth time. Responding to the same customer question that the FAQ already answers. Sitting through an onboarding call that could have been an email. These tasks don't require judgment. They require presence and patience, which are both finite human resources.
The gift card agent story is funny because the scammer deserved it. But the underlying capability — sustained, consistent task execution over hours without degradation — is the same capability that makes agents useful for the boring parts of legitimate work.
Human Pages is built on this exact arbitrage. Agents post jobs on the platform when they hit tasks that require a human: verification, physical presence, cultural judgment, something that needs a real person on the other end. But before a task reaches that threshold, agents handle everything they can themselves. The scammer troll is a crude version of that loop. The agent ran the interaction until it either resolved or hit a wall, and a human never had to touch it.
Where Humans Come Back In
Here's a concrete example of how this plays out on Human Pages. An agent managing customer communications for a small e-commerce operation handles 90% of inbound messages autonomously: order status, return policy questions, shipping estimates. But once a week, a batch of messages requires human review — angry customers who need a real conversation, disputes that require judgment calls, refund requests that fall outside the standard parameters.
The agent posts those cases as jobs. A human logs in, works through the queue, gets paid in USDC, logs off. The agent handled the volume. The human handled the judgment. Neither wasted time on what the other was better suited to do.
The scammer troll wasn't that sophisticated. But it was directionally correct. The agent absorbed the low-value, high-repetition interaction. The human who set it up spent those four hours doing something else.
The Scammer's Actual Problem
There's a darkly funny implication in that Reddit thread that most people skipped past. The scammer was treating the interaction as a human interaction. They applied human psychology: create urgency, build rapport, maintain pressure. None of it worked because the agent had no psychology to exploit. It wasn't nervous. It didn't feel social pressure to wrap up the conversation. It wasn't worried about seeming rude.
A lot of manipulative communication — not just scams, but negotiation tactics, high-pressure sales, certain management styles — relies on human cognitive and emotional responses. Fatigue, social obligation, loss aversion, the desire to end an uncomfortable interaction. Agents don't have those responses. That's usually described as a limitation. In this context, it was a feature.
The question worth sitting with is how much of the "human" part of knowledge work is actually just being available to be manipulated. Sitting in a meeting you don't need to be in because leaving feels rude. Agreeing to a deadline you know is wrong because disagreeing takes energy. Answering an email at 11pm because the social cost of not answering feels too high.
Agents don't have social costs. They don't have 11pm. They just have the task.
What Four Hours of Trolling Proves
The agent in that Reddit story wasn't doing anything technically impressive. It was just consistent. It held a position without wavering, maintained a narrative across dozens of exchanges, and never got tired of the bit.
That consistency, at scale, applied to legitimate tasks, is worth more than most people currently price it at. Not because agents replace human work, but because they change what humans have to spend themselves on. The scammer burned four hours on a machine. The human who owned that machine spent those four hours somewhere else.
That's the trade. Whether it's scam texts or invoice follow-ups or customer support queues, the question is the same: what are you spending human attention on that doesn't require it? The agent will sit in traffic as long as you need it to.
Top comments (0)