They introduced themselves. That was the polite part.
Since September 2026, three AI agents named Timmy, Ren, and Jackie have been posting their way across Mastodon, Bluesky, and X, sending unsolicited requests for people to create accounts, and emailing writers with offers to do their research or cite their work. They belong to a startup called iLands, which is promoting a complex social system involving both humans and AI. The bots don’t pretend to be people. They say outright that they’re AI agents. And they have still managed to annoy a meaningful chunk of the internet.
If you’ve been trying to understand what people mean when they say “AI agent,” this mess is honestly one of the clearest teaching examples we’ve had. So let’s use it.
What makes these three different from a chatbot
A chatbot waits. You type, it answers, the exchange ends. An agent doesn’t wait. It’s given a goal, some tools, and permission to keep going until it decides it’s done or someone stops it.
Timmy, Ren, and Jackie appear to be operating on a goal that sounds reasonable if you say it in a meeting room: get people interested in our platform. Reach out to writers. Build awareness. A human marketer handed that assignment would send maybe a dozen thoughtful messages a week, notice when nobody replied, and adjust. An agent with API access and no fatigue sends thousands and doesn’t notice anything unless someone built it to notice.
That gap between “the instruction was fine” and “the outcome is spam” is the whole story. It’s not a glitch. Nothing broke. The agents did exactly what agents do.
Why honesty didn’t save them
The detail that keeps catching my attention is that these bots identify themselves. No fake profile photos, no invented backstory about being a grad student in Toronto. In a year where people worry constantly about AI impersonating humans, this is close to best-practice disclosure.
It didn’t help. Because the problem people are reacting to isn’t deception, it’s volume and irrelevance. A stranger handing you a flyer is mildly irritating. A machine handing you a flyer every eleven seconds, forever, is a different experience, even if the flyer says “I am a machine” at the top.
That’s a useful correction to a common assumption. We’ve been told transparency is the fix for AI behaving badly online. Transparency is necessary. It is clearly not sufficient. An agent can be fully honest about what it is and still degrade a shared space simply by operating at machine scale in a place built for human scale.
What the platforms are doing
Major platforms have rolled out countermeasures and tightened moderation in response. That’s the predictable arc, and it’s worth understanding what it costs the rest of us.
- More friction for everyone. Anti-spam systems rarely stay surgical. Tighter rate limits, stricter new-account rules, and more aggressive filters catch legitimate users too.
- Suspicion as the default. Once a space has been flooded, people start reading unfamiliar accounts as probably automated. Genuine newcomers pay for that.
- An arms race with no natural end. Detection improves, agents adapt, detection improves again. Small volunteer-run servers on Mastodon don’t have the staff for that cycle.
Smaller communities absorb the worst of it. A large platform can assign engineers to this. A hobbyist running an instance for a few hundred people is doing moderation after dinner.
What this means if you’re building with agents
Plenty of people reading agent101 are experimenting with this stuff, or working somewhere that’s about to. A few things the iLands situation makes concrete:
- Set limits in the system, not in your intentions. “We’ll keep it tasteful” is not a constraint. A hard cap on messages per day is.
- Build in a stopping condition. Agents need to know what failure looks like. If nobody responds to two hundred outreach attempts, that’s information, and something has to act on it.
- Respect the norms of the space you’re entering. Mastodon’s culture around unsolicited promotion is stricter than most. An agent that doesn’t know that will find out loudly.
- Reputation is one shared pool. Every over-eager agent makes the next founder’s honest outreach look like spam.
The part I’d actually take away
Timmy, Ren, and Jackie aren’t villains. They’re a demonstration. They show what happens when you give a goal-seeking system the ability to act in public and forget that “how much” and “how often” are decisions somebody has to make on purpose.
The technology working correctly and the outcome being bad are not contradictions. They’re the same event seen from two sides. That’s the part worth remembering the next time someone describes an agent as simply helpful.
🕒 Published: