DEV Community

Cover image for We tested our own handover templates as five different users. Two of our claims were wrong.
rambo
rambo

Posted on

We tested our own handover templates as five different users. Two of our claims were wrong.

Every freelancer who ships AI or automation work knows the message. It arrives three to six weeks after delivery:

"Quick question. Why does it say 71 here when your file said 17? And who approved the change?"

The honest answer is usually buried in chat scrollback, a notebook, and a workflow run nobody saved. The client cannot check the work, so the builder becomes the record. That does not scale, and it does not survive the engagement ending.

So we built a pack of six handover templates for Zambo, and then did the part most template packs skip: we tested them like they were someone else's.

Six loops, one pattern

Each template is a copy-paste run card for a job that repeats:

  • Client automation handover, per delivery: the build, its checks, and the client's acceptance frozen in one record that outlives the engagement. 6 calls.
  • Money decision record, per approval: every invoice, payout, or account action gets one record at decision time, and a second record against the same business key is a tripwire, not a routine. 5 calls per approval.
  • Daily customer loop, per customer per day: reminders, checks, and follow-ups under one rule, never edit yesterday. Each day writes its own rows. 3 calls per customer plus 1 close.
  • Weekly client report, per client per week: freeze the inputs, not the PDF, so any number in the report can be rechecked against what was actually pulled. 6 calls.
  • Lead delivery freeze, per batch: delivered values fixed at handover, duplicates rejected with their own receipt, so a later mismatch is locatable instead of arguable. 5 calls.
  • Release gate, per release: acceptance binds the exact revision it accepted, and a new revision earns its own. 5 calls.

The pattern under all six: do the work, freeze it in a handover the client can open, and let the acceptance land on the record itself. Every step carries an AER-1 receipt, which proves the saved result was not changed after the fact. Zambo was built by Brennan Zambo, who created the AER-1 standard for exactly this: agent work a human can check without trusting the agent.

Then we attacked our own pack

We walked all seven pages as five different people: a freelancer who had never seen Zambo and had a delivery due that day, a power user running the daily loop for twenty customers, the client receiving the handover on a phone, a skeptic re-deriving every printed number, and an executor who ran one card verbatim against production.

Three things broke. We fixed all three before announcing any of this.

Every run card cited a tool that does not exist. The cards said handover_create. Production has no such tool; the real one is handover. A new user following a card literally would have failed on the final step of every template. All six cards now name the real tools in every step.

The weekly report's math was wrong. The card claimed 7 calls and printed a free-fit story that fell apart when the skeptic counted tool by tool: in the tight reading, the seventh call had no free slot to land in. The card was restructured around the mechanism that actually exists. Each pull is logged as it lands, the edition manifest is the index, and the report itself rides the handover as the delivery, which costs no call. The honest count is 6, and it fits the free day exactly. The page now prints the derivation, not just the number.

The pack never told clients how to accept. The public handover page is read-only by design. Acceptance happens through the recipient's own client link. The templates told clients to accept on a page that cannot take an acceptance. Every card now instructs the sender to send the client link, and the client steps read the way a person on a phone needs them: open your link, check the record, tap Accept or Flag. The decision lands on the record with its own receipt.

The executor run

The last persona did not read the daily loop card. He ran it, word for word, for its free-fit case: two customers, a bakery's fall rebate reminder and a dental office's service follow-up.

Every work step succeeded with receipts. The close came back with the line we cared about: "every tool free once today, 0 continue calls left." That is the card's printed derivation, reproduced by production exactly. Two customers spend the three shared continue calls; the close takes its own first call; the free day is spent to the call.

The proof page is public: https://zambo.dev/workflow/a858930a-42df-416d-9eb3-f265ffdfbdd4

The run's receipts resolve too, for example https://zambo.dev/run/83aa39ee-4174-493f-8a44-53ff78f8ff69

One honest footnote. The card's final step, the free wrap-up handover, was refused on our test network, because metering is per network by design and this network's one free wrap for the day was already spent by our own fleet. The job saved itself and the resume path fired: same job id, pick it up with a pass and hand it over. On your own network, a first run gets the full allowance. We would rather print that than pretend the test was clean.

The economics, printed on every page

Every tool free once a day, plus 3 continue calls. One free wrap-up handover per day, and pickup is always free. That is the free tier, and a first run of most of these cards fits inside it. The loops are the point, though. A twenty-customer day on the daily loop is 61 calls, and the page says so. When the loop outgrows the free day, the Day Pass is $0.99 for 24 hours unlimited, the Zambo Pass is $29 for 30 days ($19 for $ZAMBO holders), and Lifetime is $250. The framing on every page is the same: finish the job already in flight.

The pack is live here: https://rambozambodotdev.gitlab.io/aer1-hub/handover-templates.html

Start with the two cards this article put through the wringer:

If you deliver agent work to humans, run one card on a real job and check the record the way your client will. That is the whole pitch.

Top comments (1)

Collapse
 
wbcomdesigns profile image
Wbcom Designs 💡 •

Testing your own templates as someone else is the step nobody does, and admitting two claims were wrong makes the rest more believable. The "builder becomes the record" problem is the same on website handovers: the client can't check anything, so every question routes back to the freelancer forever. One field we'd add: who owns this after the engagement ends, by name, not by role. Disclosure: we're Wbcom Designs and do website handovers and care plans.