A shift is ending on a children’s ward. The resident going home works through the patient list for the one staying: how sick each child is, a short summary, the jobs still to do, what might go wrong and what to do if it does.
A handoff could easily end there. Under I-PASS, the routine Boston Children’s Hospital pediatricians Christopher Landrigan and Amy Starmer built, it isn’t over yet. The last letter stands for synthesis by receiver. The doctor staying has to talk now, summing up what they just heard, back to the person who said it.
At the end of a long shift, one more step looked like a cost nobody needed. “Fifteen years ago, there was much more resistance,” Starmer said in 2022. But when the program went into the pediatric residency programs of nine hospitals, observers timing the doctors found it did not add time to handoffs. Across 10,740 admissions, preventable adverse events, errors that actually hurt a child, fell 30%, from 4.7 to 3.3 per 100 admissions.
That result belongs to the whole program: training, a written handoff tool, supervisors watching. Still, the last step does a job nothing else in it can. The doctor leaving knows what they meant. Only the doctor staying knows what arrived.
Have people tell you what they’ll do and why. The gaps show up before the work starts.
The church with the spire
A friend gives you directions: left at the church. You say “left at the church,” and she nods. Nothing has been checked; her words have come back to her. Say instead “left at the one with the spire,” and she can tell you it’s the other church.
That is the difference between a read-back and a backbrief. A read-back proves the words arrived. A backbrief shows what the listener made of them, which is where the trouble usually is. The speaker can’t find it alone: what they already know fills every gap as they talk. That is the curse of knowledge.
The I-PASS researchers didn’t only check whether the summary happened. In a 2022 follow-up across 32 hospitals, adult and community hospitals among them, they measured how often it was good. Before I-PASS, incoming doctors gave a high-quality synthesis 31% of the time. Afterward, 83%.
What comes back up
The military historian Stephen Bungay uses the term for something bigger than a handoff. In a Harvard Business Review article published in 2011, the year of his book The Art of Action, a manager briefs his group on what it is expected to do and why, and the group works out what that means for it. Then comes the “back-briefing”: the group’s leader presents the output to the people he or she reports to.
Bungay says three things happen there. The group checks its understanding of the direction it received. The bosses “gain clarity about the implications of the direction they originally gave and may revise it.” And if several teams back-brief together, the boss can check for “gaps, overlaps, and coherence.”
The second one is easy to miss. In his fictional example, an engineer named Joe and his team come up with metrics that differ from the official targets, and the back-brief is where that gets discussed. Bungay guesses Joe’s bosses would probably revise the metrics. A parrot can’t tell you that your instructions were wrong. A backbrief can.
He traces the briefing method to the 19th-century Prussian army and says it is now part of military practice throughout NATO. The point of the whole method is the one made on the commander’s intent page: people who know why can still act when the plan stops fitting. The backbrief is how you find out whether they know.
Forty people, early June
Say you ask a new colleague to book a venue for the team offsite: forty people, early June, somewhere near the office. “Got it,” he says, and that is the read-back.
Ask him instead how he plans to go about it. “I’ll look for a room that seats forty around one big table, within walking distance, the first week of June.” Now you can hear two things. You wanted breakout rooms, because the day is mostly small-group work. And the first week of June is the release. Without that minute, either one might have surfaced after the deposit was paid.
The plan before the search
Some research agents build the same step in. When Google launched Deep Research in Gemini in December 2024, it described the flow like this: “After you enter your question, it creates a multi-step research plan for you to either revise or approve.” Only after you approve does it spend a few minutes searching and reading.
Google’s developer documentation makes it a setting it calls collaborative planning. With it switched on, the agent “returns a proposed research plan instead of executing immediately,” and you can review, modify or approve it over several turns. Leave it off, and the agent goes straight to work.
The parrot problem carries over. A plan that restates your question as bullet points is a read-back with formatting. What’s worth reading is what the agent chose: which years, which kinds of sources, what it quietly left out. Bungay’s second point holds too. Reading the plan, you may find that the question you typed is not the one you meant.
One thing differs. On the ward, the doctor leaving hears the summary and can say “no, the other one.” With an agent, you are the only one who knows what you meant. The plan sits there whether you read it or not.
The resident going home doesn’t leave when they stop talking. They leave after the one staying has spoken.