It arrives about three weeks after delivery, usually on a Tuesday:
it stopped working.
If you build automations or AI workflows for clients, you know this message is not one question. It is three, wearing a trench coat:
- What state was this thing in when you handed it over?
- What changed since?
- Whose accounts, keys, and connections is it actually running on?
Answer from memory and you get a reconstruction. The client disputes the reconstruction, because their memory says something else, and now you are doing an unpaid rebuild to end an argument about the past. The fix is not better memory. The fix is that the delivery itself left a record the client already holds, written at the moment of handover, that neither of you can quietly edit afterward.
What a delivery record actually holds
When a job is wrapped as a handover on Zambo, the wrap is refused unless it carries four things: a summary of what was done, the goal it served, the named recipient, and the next steps. Around those, the record keeps the state of the job at sign-off and every step that produced it, each step with its own AER-1 receipt. A receipt proves the saved result was not changed after the fact; it does not prove the result was right, and the record says so on its face. Acceptance is the client's recorded decision, and it moves only when the client moves it.
A live one from this morning, so you can see the shape instead of trusting my description: https://zambo.dev/handover/87a00e2c-20c5-4467-81d2-a6811b0499b6, opened Nov 19, 2026, with the job summary, the pending client approval, the dispute pack, and the pickup prompt the next agent uses to continue the same job in a different AI.
This is not a small pilot. As of this afternoon the public counter at https://zambo.dev/api/stats reads 150,732 receipts, 2,208 of them today. The records are the normal exhaust of agents doing work, not a demo someone staged.
The worked example: one "it stopped working," answered in one link
Say you delivered a weekly reporting automation: it pulls a client's sales export every Monday, cleans it, and posts a summary to their channel. You wrapped the delivery as a handover. The client accepted it. The record shows the state at sign-off, including the part everyone forgets to write down: the automation runs on the client's own workspace connection, under their account, not yours.
Three weeks later: "it stopped working."
Without the record, your next hour is archaeology. Which version did they accept? Did someone on their team reconnect the workspace in week two? Was it ever running on your credentials, so your key rotation killed it? Every answer is a claim, and the client can counter each one with a claim of their own.
With the record, you open the same link the client already has:
- State at sign-off: accepted, with the exact configuration that was live when they accepted it. No "I think we left it pointing at the test sheet." It is written there.
- What changed since: every continuation of the job is a new record chained to the same handover, so a week-two reconnect by their teammate is a dated entry, not a rumor. If nothing changed on the job, the record shows that too, and the absence is itself a finding: the break is upstream of the automation, in the thing feeding it.
- Whose accounts: the record names the owner of the connection it runs on. If it runs on theirs and their admin revoked access, the fix is a two-minute re-grant on their side, and the record is what lets you say that without it sounding like blame.
The reply you send is one link and three sentences. The fix, when there is one, starts from a shared fact instead of two competing memories.
Why this is the whole business model, not a nice-to-have
The pre-delivery proof pack matters: it gets you paid without an argument at invoice time. But the money leak for freelancers and small agencies is not the invoice. It is the month after: the state questions, the "quick calls," the rebuilds you eat because you cannot prove what you left behind. Every one of those calls is the same three questions. A living delivery record is the only version of the answer that gets cheaper every time you use it, because the client can open it without you.
The economics on my side are deliberately boring. The free tier includes one wrap-up handover per day, and pickup is always free, so a single delivery a day costs nothing to record. When delivery day means several handovers at once, the Day Pass is $0.99 for 24 hours of unlimited use, wrap-ups included: https://zambo.dev/day-pass/. Priced against one avoided rebuild argument, it is not a close call.
Set it up before the next delivery, not after the next argument
- Connect your AI to https://zambo.dev/mcp (one URL, no install per tool).
- Do the job the way you already do it.
- Wrap the delivery as a handover: summary, goal, the client as recipient, next steps, and the ownership facts you would otherwise forget by week three (whose account, whose key, what it depends on).
- Send the client the handover link. Their acceptance is recorded on it.
- When the message comes, and it will, answer from the record.
Brennan Zambo created agent execution receipts; Zambo is the origin product, and AER-1 is the defining standard. The standard is open and other tools implement it, which is the point: a delivery record your client can check without you only works if checking it does not require trusting you.
Deliver the work. Leave the record. Let the record take the Tuesday meeting.


