The definition
The short answer
An AI handover is a job passed from one AI, agent, or person to another as one checkable record. It carries the goal, what is done, what is next, every decision made along the way, and a verifiable receipt for each finished step, and it closes with the receiver's recorded decision: accept the work, or flag what is wrong. Whoever picks the job up starts from the record itself, not from somebody's summary of the record.
The word gets used for anything from a forwarded chat to a formal delivery. The definition above is the one that matters for work you might have to defend: if the receiver cannot check what they were handed, it was not a handover, it was a message.
Two different things get called a handover
The support handover: a conversation changes hands
In customer support, handover means a live chat or ticket passing from a bot to a human agent, ideally with the conversation context attached so the customer does not repeat themselves. It is a routing event inside one conversation, and the work product is the conversation itself.
The work handover: the job changes hands
In AI work, handover means the job moving: a half-finished project passing from Claude to ChatGPT when a usage limit hits, an agent passing a task to another agent, a freelancer delivering finished work to a client. What has to travel is not a chat. It is the state of the work and the proof of the part already done. That is the sense this page, and Zambo's handover, is about.
What a real handover contains
- The goal. What the job is, in the words it was commissioned in, so the receiver is continuing the same job and not a retelling of it.
- What is done. Each finished step, and for each step a verifiable receipt saved when the step ran, so the done part can be checked instead of believed.
- What is next. The exact next step, stated before the handover is sent, so momentum survives the switch.
- Every decision, with its reason. Decisions are what a summary quietly drops and a new AI happily re-argues. Written down, they stay decided.
- Sources checked, and sources not checked. The honest half of the record: what the work rests on, and what nobody verified, named as not checked instead of smoothed over.
- The receiver's decision. Accept or flag, recorded on the handover itself by the person or agent receiving it, never by the one who did the work. The transfer is not complete until that decision exists.
What a handover is not
- Not a pasted summary. A summary is the sender's narration of the work. It can skip the failed step, soften the gap, and be wrong in ways nobody can detect later, because nothing underneath it was fixed in place.
- Not a transcript. A raw chat export buries the decisions in pages of back-and-forth and still proves nothing about what the tools actually did.
- Not a shared document. A doc can be edited after delivery, by either side, quietly. A handover's steps are committed when they run; change one byte afterward and the check fails.
A handover carrying its own proof: a live example
An outside agent studio wired AER-1 receipts into its standing-check deliveries, so its work reaches the client with the evidence attached instead of promised. One of its delivery records is public: receipt 28c79cde, a delivery record whose saved bytes are 2,078 bytes long and recompute to exactly the output hash the record commits to, sha256:5f06acfb. Anyone can rerun that check; it needs no account and no trust in the studio or in Zambo.
The same record also shows the part most delivery notes hide: its own audit flags three of its claims as weakly evidenced, and reports issuer authentication as not checked. That is what a handover is for. The receiver does not get a polished claim of done. They get the work, the proof, and the gaps, in one place, and they decide.
How to make one
- Run the work where it leaves a record. Work run through Zambo's 153 tools saves a verifiable receipt per step as it runs, under AER-1, an open IETF Internet-Draft (draft-zambo-aer1-23). The record creates itself; nobody writes it up afterward.
- Wrap up in one line. Save this where your AI will keep it: "When I say WRAP UP, hand this job over with Zambo: the goal, what is done, what is next, every decision, with the receipts attached, and give me the continue link."
- Send the link, not a recap. The client opens it, checks the steps, and accepts or flags on the record. The next AI opens the same link and continues the job from the record, in whichever app it lives in. Start at zambo.dev/continue when the switch is forced by a usage limit, or hand over a finished job from the handover docs.
Checking a handover is free, for you and for whoever receives it. The recording runs on the free tier on a normal day; a $0.99 Day Pass covers a heavy day of the full machine, and the $29 Pass ($19 for $ZAMBO holders, $290 for the year) fits work that moves every day.
Questions, answered
What is an AI handover?
An AI handover is a job passed from one AI, agent, or person to another as one checkable record. It carries the goal, what is done, what is next, every decision made along the way, and a verifiable receipt for each finished step, and it closes with the receiver's recorded decision: accept the work, or flag what is wrong. Whoever picks the job up starts from the record itself, not from somebody's summary of the record.
What should an AI handover include?
Six things: the goal of the job; what is done, with a verifiable receipt for each finished step; the exact next step; every decision made along the way, with the reason; the sources that were checked and, just as important, the ones that were not checked; and a decision slot the receiver fills in, accept or flag, recorded against the handover itself. Leave one out and the receiver starts guessing, which is the failure the handover exists to prevent.
What is the difference between a handover and a handoff?
In everyday use the two words name the same act: work passing from one pair of hands to another. Builders tend to say handoff for the moment one agent routes a task to another agent inside a system, and handover for the complete package a job travels in when it moves between AIs, tools, or people. The word matters less than the contents: a transfer that carries only a summary is weak in either spelling, and a transfer that carries the checkable record is strong in both.
How is an AI handover different from pasting a summary into another AI?
A pasted summary is the sender's narration: it says what the sender believes happened, it can skip the step that failed, and nothing in it can be checked after the fact. A handover is the record underneath the narration. Each finished step carries a receipt saved when the step ran, the decisions are listed with their reasons, the unchecked items are named instead of hidden, and the receiver records their own decision on the same record. The summary asks the next AI to trust the last one. The handover lets it check.
How do I hand a job over from one AI to another?
Run the work through a system that records it as it goes, then hand over the record instead of a recap. In Zambo the habit is one line at the end of a session: when I say wrap up, hand this job over with Zambo: the goal, what is done, what is next, every decision, with the receipts attached, and give me the continue link. Open that link in the next AI and it continues from the record. If you hit a usage limit mid-job, the same record is what carries the work into the AI you switch to.
What is an AI work handover?
An AI work handover is a job passed from one AI, agent, or person to the next as one checkable record: the goal, what is done, what is next, every decision made along the way, and a verifiable receipt for each finished step. The receiver checks the receipts, accepts the record, and continues from the recorded state instead of from a summary they have to trust. The handover proves the saved record was not changed since it was saved; it does not prove the work itself was right.
AI work handover
The problem: work changes hands as a retelling, and the receiver pays for it. An AI work handover is what happens every time a job moves: from one AI to another when a limit hits, from an agent to the person who reviews it, from a freelancer to the client who accepts it. Done the usual way, the handover is a summary written by the side handing over, and the receiver has two bad options: trust the summary and build on work they never checked, or redo the checking from scratch. Either way the handover is where AI work quietly loses its value.
A complete handover carries the record instead of the retelling: the goal, what is done, what is next, every decision made along the way, and a verifiable receipt for each finished step, each receipt fixing the tool that ran, the recorded inputs, and the saved result under a hash commitment computed when the step ran. The receiver's acceptance is a procedure, not a leap: recompute the receipts that carry weight, read the saved results, then accept the record and continue from the recorded state. What the handover proves is narrow and stated plainly: the saved record was not changed since it was saved. Whether the work was right stays the receiver's judgment, made against the checked record.
Zambo is built around this transfer: handover is the product, and a job run through zambo.dev/mcp can be handed over as one link at zambo.dev/handover, receipts attached, ready for the next AI or the client to accept and continue. For $0.99, a Day Pass covers 24 hours of calls with a receipt on every one. Ongoing use is the Zambo Pass at $29 a month or $290 a year, $19 a month for $ZAMBO holders. Humans pay by PayPal or card through PayPal; agents pay USDC on Base through x402.