Bystander Effect

Here is one question in one Slack channel, and everyone who saw it. The channel and the people are invented for this post, though the question is the one from the example on our homepage: "Going with seat-based?"

Three weeks later, the checkout ticket was still blocked and nobody had decided anything. Nobody in the table was careless. Every one of them saw the question, most of them cared about the answer, and each of them had a perfectly reasonable belief that it was somebody else's to resolve.

Psychologists have a name for that table. The bystander effect is the finding that a person is less likely to act on a problem, and slower to act, when they know other people are aware of the same problem, because the responsibility for acting is spread across everyone who knows.

An intercom and a seizure

In 1968, John Darley of New York University and Bibb Latané of Columbia published "Bystander Intervention in Emergencies: Diffusion of Responsibility" in the Journal of Personality and Social Psychology. Their setup was simple and a little cruel. Seventy-two New York University students were told they would be discussing personal problems over an intercom, each in a separate room so nobody would feel embarrassed. Partway through, one voice, a recording, began to have what sounded like a seizure.

The variable that mattered was how many other people each student believed was listening. (The researchers also varied who the other listeners were, and found that made no difference.) Some thought they were the only one. Some thought one other person could hear. Some thought four others could.

The results are in the paper's first table:

Counting reports made after the seizure ended too, everyone who believed they were alone reported it eventually. Among those who believed four others were listening, even that later count reached only 62%.

Darley and Latané gave three reasons a crowd slows people down. "The responsibility for helping was diffused among the observers; there was also diffusion of any potential blame for not taking action; and finally, it was possible that somebody, unperceived, had already initiated helping action." When one person is present, "any pressure to intervene focuses uniquely on him." With several, "the responsibility for intervention is shared among all the onlookers and is not unique to any one."

The finding that matters most is about the people who didn't act. They weren't indifferent. The researchers wrote that those students "showed few signs of the apathy and indifference thought to characterize 'unresponsive bystanders.'" Many asked whether the victim was being taken care of, and many "had trembling hands and sweating palms." Their conclusion was that the explanation "may lie more in the bystander's response to other observers than in presumed personality deficiencies."

The paper opens with the 1964 murder of Kitty Genovese in New York and a newspaper report of 38 witnesses who did nothing. Later reporting, including the New York Times's own, found that account exaggerated how many people saw the attack and what they understood. The experiment doesn't depend on the story. It was built to test the idea in a room where the numbers could be controlled.

A pending decision is an emergency with an audience

A blocked decision isn't a seizure, but the three mechanisms carry over with almost nothing lost.

Responsibility spreads. A question asked in a channel with forty people in it has forty possible owners, which in practice means none. Each person's share of the responsibility shrinks as the channel grows.

Blame spreads. When the launch slips, it will be hard to say whose fault it was that nobody made the pricing call. Everybody saw it. That makes not acting cheap.

Somebody might already be on it. This is the mechanism that modern tools amplify. In the experiment, students couldn't see each other, so they could imagine someone else had gone for help. In Slack, the signals are real but ambiguous. An eyes reaction, a "+1," a reply with numbers: each one looks like someone is handling it, and none of them is a decision. Darley and Latané also noted the role of walls that kept bystanders from communicating. A product organization has plenty of those: the Slack thread, the Confluence spec, the Jira ticket and the merge request are four rooms, and the people in each can't see what the others did.

The people in the table at the top were not apathetic either. They were busy, they were paying attention, and they were responding to each other.

Agents add bystanders

Coding agents make the room bigger. Our essay on the modern product development lifecycle describes the shift: coding agents sped up Build, and the work now piles up at the same review gates, waiting on people. Every agent that runs into an open question is one more observer of it. It notes the question in a pull request, asks in a thread, or leaves a comment on the ticket. Each of those is a new sighting, and none of them is a decision. The more agents a team runs, the more often the same open question is seen, mentioned and left where it was.

The opposite failure is worse. An agent with nobody to hand the question to can answer it itself, by building per-seat pricing because the ticket said so. Then the decision gets made, by default, by whoever happened to be writing the code.

Where Brief fits: one name, one queue

The experiment's lesson is that people act when the responsibility is clearly theirs. Brief is the product strategy agent, and a large part of what it does for an open decision is make the responsibility clear: who owns it, what's waiting on it, and where it stands.

Every open decision can have one name on it. Brief keeps proposed decisions separate from recorded ones, and a proposal can name Who should decide? When someone opens a proposal assigned to somebody else, the confirmation dialog tells them before they confirm. In Slack, /propose turns the open question into a proposal, and Edit can add an owner. On the homepage, D-241, the seat-based pricing decision, is "owned by Dana Ruiz." That's the difference between forty people who saw the question and one person whose question it is.

The question "who decides?" has an answer. On the homepage, an engineer asks on the blocked Jira ticket why it can't ship, and Brief answers there with the decisions that apply, including the pending one. When the engineer asks who can sign off, Brief answers: "Dana Ruiz, CPTO. It's at the top of her Friday brief." Nobody has to post into the channel and hope.

The pressure lands on one person. Darley and Latané found that a lone bystander acts because "any pressure to intervene focuses uniquely on him." The homepage's example of the CPO's weekly brief does exactly that: it leads with D-241 under Needs your decision, says engineering "is holding the change until you approve," and shows the three launches slipping while they wait. With Proactive Brief turned on, when reviewable items have piled up, Brief can invite you to review them in chat or send a reminder in Slack or Teams, capped at two interruptions a week. The queue reaches the person who can clear it, without depending on the crowd to escalate.

Agents report to the record, not the crowd. Coding agents set up with Brief call brief_ask before they write code. In the homepage example, an agent asked to add per-seat pricing learns that D-241 is pending, holds the change, and flags a related security review. It doesn't add one more comment to the thread, and it doesn't answer the pricing question by building it. Agents can call brief_list_pending_decisions to see the queue of decisions awaiting confirmation, and the homepage lists the checkout team's Claude Code agent, holding the pricing change, among the people and agents working from D-241.

A name on a decision doesn't make anyone decide faster, and a reminder is easy to ignore. The deciding is still a person's job. What the 1968 experiment shows is that most failures to act aren't about the person at all. They come from not knowing it was theirs. That part can be fixed by putting one name on the decision, making the waiting visible to that person, and keeping everyone else, people and agents, from mistaking a sighting for a response.

How many open decisions in your channels right now have been seen by everyone, and are owned by no one?

Objections