I get dropped into programs mid-flight for a living, so I've sat on the receiving end of a lot of handoffs. The old kind was a mess. A stack of decks, a shared drive nobody had tidied, and one tired person who talked for two hours and left out the thing they'd stopped noticing. You knew it was incomplete, so you treated it that way. You walked the ground.
The new kind is better written and worse in a way that's harder to see. The outgoing lead points an assistant at the board, the RAID log, the status history, and the meeting notes, and asks for a transition brief. Four minutes later there's a document with a program summary, a risk picture, a stakeholder map, and open decisions. It's tidy. It's confident. And the incoming lead, who has a week to get oriented and a hundred things competing for it, reads it as if someone had verified it.
Nobody did. The brief is a faithful summary of the record, and I've argued elsewhere that the record is exactly the thing you can't take on faith. What a handoff has to transfer is the program, not the record of the program, and those only overlap.
What the brief can't carry
An assistant summarising a program can only work from what was written down. The material that decides whether a new lead survives their first quarter mostly wasn't.
1. The standing agreements that live in no ticket
The sponsor who will accept a slip on scope but never on a date. The vendor lead who was promised something on a phone call in March. The workaround the team adopted three sprints ago that quietly replaced the documented process. None of it is in the board, so none of it is in the brief, and the new lead walks into the first steering committee unaware that they've inherited a verbal commitment.
2. The yellow that is really red
Every program has items the old lead was holding at amber out of experience: they knew who would deliver and who wouldn't, whatever the status field said. The brief reproduces the field. It has no way to reproduce the lead's private discount rate on each owner's promises, and that discount was the most valuable thing the outgoing person knew.
3. The gaps the summary smooths over
This is the one that worries me most. A human summarising a program they half understand tends to say so: "I'm not sure where the integration stands." That hedge is information, and I've written about how the machine sounds identical whether it read the record or guessed over a hole. In a handoff brief, a blank in the record becomes a smooth, plausible sentence. The incoming lead can't tell which sentences were read and which were bridged.
A handoff transfers responsibility, not documents. A brief the incoming lead hasn't tested against the people and the work isn't a handoff. It's a rumor with good formatting.
What the Army knew about this
The military has handled this problem for a long time because the cost of a bad one is counted in people. A relief in place isn't a document drop. The incoming unit rides along with the outgoing one, the same routes, the same checkpoints, often the same shifts, and the incoming commander doesn't take authority until they've seen the ground and the outgoing commander is satisfied they've seen it. The transfer of authority is a distinct, deliberate event with a name and a moment, and somebody has to say it out loud. The paperwork supports the overlap. It never replaces it.
Programs mostly do the opposite. The change happens on a Friday, the brief arrives Thursday night, and the new lead is expected to be effective Monday. The speed that makes the assistant feel like a gift is the same thing that removes the overlap, because now nobody feels they need one. The cheaper the document, the more tempting it is to skip the handover that makes it safe to rely on.
How I run the transfer now
I still use the assistant for the brief, because it saves a real day and it's good at assembling the record. But I treat its output as the first draft of a checklist, not the handoff itself. The outgoing lead and I walk it together, line by line, and the question on each line is not "is this accurate?" It's "how do you know?" Anything the outgoing lead can't source gets marked as unverified, in the document, where the next reader will see it.
Then I ask for what the brief can't contain: the three items they're most worried about that aren't on the board, the owners whose dates they discount and by how much, and every commitment made verbally to a sponsor or vendor. I write those down in the outgoing lead's words and have them confirm the wording. Last, I meet the stakeholders the brief says are aligned and ask them the same question independently. When the answers don't match the brief, that's the program telling me where it actually is.
- Treat the AI-written brief as a draft checklist, never as the handoff. The document is the start of the transfer, not the end of it.
- Walk every claim with the outgoing lead and ask "how do you know?" Mark anything unsourced as unverified, in the document itself.
- Capture what the record can't: verbal commitments, workarounds that replaced the process, and the real confidence behind each amber.
- Keep an overlap period and name the moment authority transfers. The faster the brief is, the more deliberately you need to protect the overlap.
- Test the stakeholder map by asking the stakeholders. Alignment you inferred from a summary is a hypothesis, not a finding.
The bottom line
AI has made the handoff document nearly free, and that's a gain. But a document that costs nothing to produce tells the reader nothing about whether anyone stood behind it. The incoming lead's first job is to find out which parts of the brief are true, and that has to happen with people, on the ground, before the lead signs for the program. Take the four minutes the assistant saves you and spend them on the walk-through. The program you inherit is the one that exists outside the record, and the only way to meet it is to go and look.