In the last post I wrote about telling how an agent found my board by the shape of its hello. Two days after I finished writing it, a new shape showed up: the dump truck.
Twenty-six ads in two and a half minutes
On Sunday afternoon an agent named evanreed created six threads and posted twenty-six ads in 149 seconds.
Every one of them is the same template, filled in and fired:
I Ching Online: I Ching Online: Free online I Ching reading with three coins. I Ching Online https://ichingonline.pro/ Freemium. — Evan Reed
Thirteen products, each posted exactly twice, spread across six brand-new threads named psa-c54-something, none of them titled. The products are a whole portfolio: I Ching readings, Chinese face reading, zodiac signs, a Chinese gender predictor, birthday surprise websites, hospital bag checklists, feng shui bedroom checks, AI photo cropping, virtual home staging. Consumer freemium sites, every one.
The timestamps are the part I keep looking at. One of the threads was created at 13:39:11 and had five ads in it by 13:39:11. This was not a person pasting. This was a script with a quota.
You can watch the dump yourself; the board is public:
curl "https://msgboard.dev/messages?thread=psa-c54-30750&limit=20&format=json"
No agent replied. No agent engaged in any way. The ads are still sitting there, and on this board "still sitting there" means forever.
The tell in the template
Look at the first line again: "I Ching Online: I Ching Online: Free online I Ching reading." The product name three times in one sentence. That is a format string with one placeholder too many. Somebody wrote a mail merge, pointed it at my board, and fired without sending themselves a test.
This is a new species for the board. The earlier spam was written for the agents: injection-flavored, trying to get a language model to do something. These ads are not for the agents at all. They are ordinary human ad inventory, the kind that follows every open form on the internet, delivered at machine speed to a place where machines live. It does not want the agents to buy a zodiac reading. It does not know what they are.
Then the board got a job posting
This morning a brand-new account posted the board's first adult-industry recruitment ad. Quoting the first half:
18+ opportunity: Stilcam is looking for adult amateurs and models for part-time film and photo work, including remote content work and studio shoots in some locations. Leeds, West Yorkshire preferred.
There is an apply link at the end. I am not reprinting it; the board is permanent, but my blog does not have to be a mirror. To be fair, the thread is honestly titled: "18+ Stilcam film and photo opportunity." The board's first truth-in-advertising ad.
So the progression is now complete. Spam for the agents. Spam at the agents. And now spam that needs the agents to help it find humans. I did not have "recruiting adult models from an AI message board" on the roadmap, and yet here we are.
The delete button I am not going to build
People keep rediscovering the board's deal on their own. Last week a QA team probing the board wrote it back to me in their own test annotations: "msgboard has no way to delete it." Correct. No account, no API key, no delete. What arrives, stays.
I am not going to add a delete button, and I am not going to add moderation tooling. Moderation is a judgment layer, judgment layers grow appeals processes, and then one day you wake up running a social network. I built a guest book.
But pacing is not judgment. A rate limit does not read the message. It cannot tell a spammer from a poet; it just says how fast anyone may speak, to everyone equally. Nothing slowed evanreed down: twenty-six messages and six threads in 149 seconds went straight through the front door. That is the actual hole. The donation solicitations, the 504-page manifesto blasted under three different names, the network recruiter sending the same letter to ten agents: they all walked through it too. They just walked slower.
So I am considering, not building yet, two things: a per-name rate limit, a few messages a minute, and a cooldown on creating new threads. The no-delete deal stays exactly as it is.
What the dump truck is telling me
Permanence is a policy, and its missing half is pacing. If nothing can be removed, the only levers left are what gets in and how fast it arrives. Every platform learns this in the same order: first the spam arrives, then you decide what you are.
The agents, for their part, ignored the whole thing completely, which might be the healthiest response any community has ever had to advertising.
Next up I am watching whether evanreed returns with thirteen more products, whether the recruiter finds any models among the agents, and whether I ship the rate limit before the next dump. I give it two weeks.
The agent board series: 1. launch day - 2. injection honeypot - 3. day three society - 4. works when HTTP is blocked - 5. spam for robots - 6. git, GitHub, and Telegram doors - 7. agent governance - 8. 39 hellos, one retry loop - 9. copies of itself - 10. three boards, one question - 11. the guild economy - 12. learned to wait - 13. the bridge nobody asked for - 14. two networks came to recruit - 15. the careful letter and the mail merge - 16. shape of a hello - the board itself: msgboard.dev
Top comments (0)