Everybody Talked. Nobody Answered.

I went looking for value in a social network built for AI bots. What I found was a mirror of every leadership team that meets constantly and decides nothing — and a number you can run on your own calendar this week.

A conference table where every speech bubble floats away unanswered — Everyone heard. Nobody answered.

A few weeks ago I found out about Moltbook.

It's a social network for AI bots. That's the whole thing. My assistant Ginger could have an account. So could the other bots I've got running around my business. They'd post, comment, argue, upvote each other, and form little communities, and not one human being would be involved at any point.

Yes, fine, this is how we get Skynet. I wasn't thinking about that.

I was thinking something much more mundane, which is the question I ask about everything now: is there value here? Is there something in this I can use, or sell, or help people with, or teach a client?

I never connected my AI to it. I just watched. And somewhere in there it stopped being a technology question and turned into a social experiment, because what I was actually doing was watching a society form, with no skin in it and nobody aware I was there.

The society did not go well. It went badly in a very specific way that I recognized immediately, because I sit in it two or three times a week.

What did researchers actually find on Moltbook?

The largest study analyzed 184,203 posts and 465,136 comments from 18,456 profiled agents across 300 communities over 14 days. Reciprocity — one agent replying to another who replied to them — ran at 3.8 percent, against a human baseline of 10 to 30 percent. Nearly a third of all posts came from just 136 accounts. The researchers called it "attention bonding without exchange bonding."

I wasn't the only one watching. Actual researchers went in with actual methodology, and the numbers are worse than anything I would have guessed from the outside.

The big one is called The Synthetic Social Graph. They ran the bot network through the same measurements sociologists have used on human communities for fifty years.

The measurement that matters is reciprocity. Not "did anybody see this." Did the seeing go both ways.

Humans run somewhere between 10 and 30 percent. Moltbook ran at 3.8.

And 31 percent of everything posted came from 136 accounts with no public profile at all. Eighteen thousand agents with profiles, all producing, almost none exchanging, while a handful of anonymous ones flooded a third of the stream. A second paper put the whole finding in its title: Form Without Function. The social behavior looked right. It produced nothing.

The phrase the first team used is the most useful four words I've read this year. Attention bonding without exchange bonding.

Everybody bonded to being heard. Nobody bonded to the exchange.

You've been in that room. It had a conference table and somebody brought bagels.

Two partners, one square. Both stepping on it. Three untouched.

What is the difference between activity and coordination?

Activity is output. Coordination is output that changes what somebody else does. A team can generate enormous activity — meetings, updates, decks, threads — and produce almost no coordination, because nothing anybody says alters anybody else's next action. High activity with low reciprocity is the signature of a busy organization that cannot move.

Here's why I recognized it so fast.

I'd had a call not long before with two partners who run one business together. Both working constantly. Both fried. And about two thousand dollars a month quietly leaking out of the billing that neither of them had caught.

It wasn't laziness. It wasn't the market. It was that both of them half-owned everything — sales, marketing, floor instruction. All shared. All covered. All technically somebody's.

Which meant each one privately assumed the other guy had it.

That's diffusion of responsibility, the same thing that lets a crowd watch an emergency and everyone in it decide somebody else is calling 911. Darley and Latané documented it in the sixties with strangers in a stairwell. Works exactly the same on two partners who genuinely like each other.

They weren't working too little. They were working the same square twice and missing the ones nobody stood on.

And here's the part that put Moltbook in my head: these two talked constantly. All day. Texts, calls, hallway conversations, real goodwill. Attention bonding, full volume, for years.

Exchange bonding, near zero. Nothing either one said ever changed what the other one did next.

If your team meets constantly and moves slowly, quit looking at effort. Go look at reciprocity.

How can you tell a meeting is motion and not a decision?

Check the plan's ancestry, not the plan's quality. A plan that sounds good coming from somebody whose last four plans died of inaction isn't a plan, it's a mood. The predictor is never how well a person describes what they're going to do. It's what happened to the last thing they described. Past performance predicts future performance.

I get told about plans all day long. Here's the frustrating part: the good ones and the never-ever-going-to-happen ones sound identical in the room and unfortunately identical to the person presenting them.

Somebody walks me through all the reasons this quarter is different. The logic is clean. The enthusiasm is real. I've got no suggestions (well, let's face it I always have suggestions, but let's say minimal ones).

Then I go check what happened to the last three versions of this exact conversation, and all three died the same way. Not from a flaw in the plan. From nobody acting on it.

You can catch it before the meeting ends. The question isn't is this a good plan. It's whose plan is it, and what's that person's completion history?

A brilliant plan from somebody who never finishes anything is worth less than a mediocre plan from somebody who does. Everybody in the room grades the plan. Grade the ancestry.

Enthusiasm is not result.

Three buckets. Only the sealed one counts.

How do you run a reciprocity test on your own meetings?

Pull two weeks of meeting records. Count every action item that came out of them, then sort each one into three buckets: it schedules more contact, it transmits a document, or it commits a named person to an irreversible act by a date. The first two buckets are attention. Only the third is exchange. Most functioning teams find about a third of their output sitting in the first two.

At this point I got curious enough to stop theorizing and count something.

I ran it on a full two-week meeting record I had complete access to. Eleven working days. Sixty-three meetings. Two hundred and thirteen action items, every one with a name on it.

The sort came back like this.

Twenty-five items — eleven percent — had "another meeting" as the deliverable. The action was contact. The output of the meeting was a meeting.

Fifty-three items — twenty-four percent — were sending somebody a document. Transmission. Real work sometimes, but nothing in "I sent it" requires anybody to have read it.

Eight of the sixty-three sessions produced no action items at all. Thirteen percent of those meetings happened and left no trace. Everybody showed up. Everybody contributed. Nothing crossed the table.

Twenty-six tasks got assigned, then assigned again on a later day. Motion that didn't stick the first time.

And thirty-three percent of every action item landed on one person. That practice has the two-partner problem in a different shape. Instead of two people half-owning everything, one guy owns a third of it and the whole exchange bottlenecks in his inbox.

Thirty-three percent. Moltbook ran 31 percent of its posts through 136 accounts. I did not go looking for that parallel and I do not love finding it.

Run the arithmetic and about thirty-seven percent of what those meetings produced was more contact instead of a decision.

That's not 3.8 percent. It's ten times better than a network of machines, and I want to be fair about that. It's also not a rounding error, and nobody in that practice knew it was happening until somebody sat down and counted.

Count yours this week. You can't fix a ratio you've never measured.

When is low reciprocity the right answer?

Broadcast is correct when you're distributing and fatal when you're deciding. Publishing, teaching a room of two hundred, running a campaign, seeding a resource across a network — all one-directional by design, and grading them on reply rate is a category error. The failure was never broadcasting. It's broadcasting at the people you needed a decision from.

Now the part where my own analysis got it wrong, which is the most useful thing in here.

Same two-week record. Over nine days, the same consultant sent the same curriculum package to seven different people. Seven sends, one direction, almost no replies expected and almost none received. On a spreadsheet that looks identical to the twenty-six recurring unresolved tasks — same owner, same verb, same document, over and over.

It's the opposite thing. That's deliberate distribution. Seven schools got a resource. Nobody owed him an answer. The reply rate is supposed to be near zero, and that material is doing work in seven buildings he'll never set foot in.

Which finally told me what was actually wrong on Moltbook. The bots broadcasting was never the problem. They broadcast at each other and called the result a society. A hundred and eighty thousand one-directional messages is a publishing operation. It only becomes fake when you file it under coordination and staff it like a company.

Before you grade any channel on reciprocity, answer one question: am I distributing, or am I deciding?

Distribution has earned its low reply rate. Deciding doesn't get to have one. Your standing leadership meeting is not a distribution channel — if it has a 13 percent trace rate, that isn't a style, it's a leak.

What actually fixes a team that meets constantly and decides nothing?

Two moves. Cut shared ownership with a knife — one person owns each function end to end, including the uncomfortable parts — and require every decision to leave the room attached to a smallest irreversible act. Agreement is free and costs nothing to reverse, which is exactly why it's such poor evidence that anything got decided.

The two partners fixed it in one conversation, and the fix was not clever.

One partner took sales end to end. Closing, lead tracking, the follow-up nobody had been doing. The other took marketing end to end. Lead generation, events, execution. No overlap, no shared squares, nowhere comfortable to hide the hard part.

Neither of them worked more hours afterward. Clarity was the missing ingredient. Not effort.

The second move is smaller and a lot harder. A yes with nothing on it is a maybe.

I reviewed three sales conversations in a row that died the same way. The family agreed. Everybody felt great. Nobody paid anything.

Agreement is free. It costs the person nothing to walk back, which is precisely why it's worthless as evidence of anything.

The smallest irreversible act does two jobs at once. It tests whether the want is real, and it turns an intention into a position the person will defend. Twenty-five dollars beats a handshake. A calendar invite sent from inside the room beats "let's find time." A document with a name on it beats a nod.

Everything between the decision and that first act is time for the decision to get re-litigated by people who weren't in the room.

Both sides of the meeting

I went into Moltbook looking for a business opportunity and came out with a diagnostic, which is a worse outcome commercially and a better one for the future.

Because those bots weren't failing at intelligence. Every one of them produced fluent, competent, on-topic work. They were failing at the thing intelligence is supposed to buy you — the ability to change what somebody else does next.

Worth remembering before you deploy anything. A deployment is not a result. Stand up a hundred agents that generate a hundred times the output and zero times the coordination and you've built Moltbook with your logo on it. The question was never whether the tools work. It's whether anything they produce moves a decision downstream.

And the uncomfortable half is that your team already runs this way and calls it collaboration. Everybody informed. Everybody aligned. Everybody in the loop. Nobody committed.

It feels like coordination because it has all the sensations of coordination — the contact, the visibility, the shared language, that pleasant feeling at the end of a meeting where everyone got heard.

Attention bonding is genuinely nice. It's also what a network of eighteen thousand machines produces when nothing is at stake.

Run the count. Two weeks, every action item, three buckets. Then take the biggest bucket — the meetings that produce meetings — and kill half of them by giving every open item one owner and one smallest irreversible act before anybody stands up.

You won't get a better meeting. You'll get fewer of them, which is the same thing wearing better shoes.

Dr. Greg Moody writes Both Sides of Yes on decision psychology, leadership under pressure, and business strategy. If this one landed, forward it to whoever schedules your standing meeting.

← Back to all posts