How do I hand an AI agent's work over to a human employee who does not use AI?
The short answer You hand over a link, not a login. When an AI agent's job is wrapped as a Zambo handover, the employee gets one web page that opens in any browser: what the job was, what got done step by step, what ha
The short answer
You hand over a link, not a login. When an AI agent's job is wrapped as a Zambo handover, the employee gets one web page that opens in any browser: what the job was, what got done step by step, what has not been checked yet, and a decision box where they accept the work or flag it, in plain language, without an account, without installing anything, and without ever opening an AI tool. The AI side of the handoff is complicated. The human side is a page they read like a delivery note.
Disclosure, once: I run ops for Zambo, the system that produces these handovers. The example below is a real, public handover you can open yourself.
Why the usual handoffs fail this employee
The standard ways of passing AI work to a non-AI colleague all smuggle in a requirement that person does not meet:
- A chat transcript asks them to scroll a conversation they were not part of and guess which parts are done, which are claims, and which still need a decision.
- A dashboard invite asks them to create an account on a tool they will use twice a year, learn its vocabulary, and find the one run that matters among hundreds.
- A summary document asks them to trust the summary. The person least equipped to audit AI work gets the artifact with the least evidence in it.
What this employee actually needs is the shape of a good paper handoff from a departing colleague: here is the job, here is what I finished, here is what I did not finish, here is what needs your decision, here is the proof for each line. That shape exists; it is just usually produced by a careful human. A handover produces it mechanically.
What the employee sees: a real handover, walked through
Open this, in any browser, signed out:
https://zambo.dev/handover/d468118b-afc1-43e5-a70c-5e198e876a04
It is a small job, an ETH price check, and everything on the page is what your employee would see:
- The goal, in one line: ETH price check. No prompt history, no system messages, just what the job was for.
- The steps, each with its state: a live_price call whose saved result reads ETH at $1,621.82 USD, and a prompt_shield check that came back with risk level none. Each step carries a verifiable receipt the employee can check on the page itself: the saved result is committed under a SHA-256 hash, and recomputing it shows whether the record changed after the fact. They do not need to understand hashing to use it; the page does the check and shows the result.
- What is not decided yet, stated plainly: the Decisions section reads "No client decision has been recorded," with approval pending. I am showing you one in this state deliberately. The page does not pretend the work was accepted, and your employee's one job on this handoff is visible at a glance: record the decision.
When they accept or flag the work, that decision is recorded on the same page, next to the steps it covers. Weeks later, nobody has to reconstruct who approved what from an email thread.
What you, the sender, have to do
The wrap is deliberately not free-form, because a handover that can be vague will be. Zambo refuses a bare wrap: a handover requires a summary, a goal, a named recipient, and next steps, and it binds in the receipts of the job's own steps, so the employee's page cannot quietly include work from a different job. A wrap attempt missing those fields comes back as a needs_input receipt instead of a page, which is the system enforcing the checklist your busiest day would skip.
Your part is then one message, in whatever channel the employee already uses: "The price-check job is done, the record is here, please check the two steps and record your decision at the bottom." That is the entire training requirement.
The honest limits
A handover proves the record of the work: each step's saved result, unchanged since it was saved, checkable by anyone. It does not prove the work was correct, and it does not make a bad job good. What it changes is who can participate in checking. The verification stops being a developer ritual and becomes something any careful employee can do from a link, which is the difference between AI work that stays inside the AI team and AI work the rest of the company can actually receive.
The receipts on these pages follow AER-1, an open format published as IETF Internet-Draft draft-zambo-aer1-26 (Informational, expires 25 April 2027; a draft, not an RFC), with a public conformance corpus of 143 vectors and 146 checks passed by independent implementers, Laurent Lemonnier at 143 of 143. The practical meaning for your employee: the page they are reading is not a proprietary report they have to take on trust; the format is public and the checks are real.
Brennan Zambo created agent execution receipts; Zambo is the origin product, and AER-1 is the defining standard.
What it costs
Reading and checking a handover costs the employee nothing, forever: pickup is always free, and receipt verification is free. On the sender side, the free tier covers a handover a day: every tool free once a day, plus 3 continue calls. One free wrap-up handover per day. Past that, a Day Pass is $0.99 for 24 hours unlimited, the Zambo Pass is $29 a month flat ($19 for $ZAMBO holders), and Lifetime is $250. The system behind this is running at a scale you can inspect: 158,096 public receipts (158K+), 2,437 minted today, read from https://zambo.dev/api/stats on November 21, 2026, at 22:18 UTC.
The test to run on Monday
Pick the next piece of AI work your team would normally hand over in a meeting. Wrap it as a handover instead, send the link to the employee who would have sat in that meeting, and ask them two questions afterward: what did the agent do, and what is waiting on you? If they can answer both from the page, the handoff worked, and it worked for the person every other method leaves behind.
Sources
- Proof handover, opened first-hand November 21, 2026: https://zambo.dev/handover/d468118b-afc1-43e5-a70c-5e198e876a04
- AER-1 corpus (143 vectors, 146 checks): https://zambo.dev/aer1/corpus.json
- Live counters: https://zambo.dev/api/stats (read November 21, 2026, 22:18 UTC)
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.