DEV Community

Cover image for How to prove AI work to a client before invoicing, with one link the client can check
rambo
rambo

Posted on Edited on

How to prove AI work to a client before invoicing, with one link the client can check

How to prove AI work to a client before invoicing

The invoice is ready. The client is not refusing to pay, exactly. They are asking the question that stalls every AI assisted delivery: what exactly did the AI do, and how do I know it did that and not something else?

If your answer is a summary you wrote, or screenshots, or a chat transcript you pasted into a doc, the client has to trust you. Trust is a slow way to get paid. A checkable record is faster, because the client stops taking your word for the work and starts verifying it themselves.

This is the flow I use. It is built on Handover, the flagship product at Zambo. Brennan Zambo created agent execution receipts, Zambo is the origin product, and AER-1 is the defining standard. A handover is one record that carries a finished job, every step receipt from that job, a named recipient, and the client decision, all behind a single link you send before the invoice conversation starts.

A real handover, opened and read

Here is a live record you can open right now: https://zambo.dev/handover/e2ef637d-663e-4020-bdc9-c69fbb1fa4de

I fetched and read this record on November 18, 2026. This is what is actually inside it, with the recipient anonymized here as the client, because the point is the shape of the record and not anyone's name:

  • The goal on the record: deliver a demo record as a checkable handover the recipient can pick up.
  • The summary on the record: every price in the job was fetched live inside the job, one call per asset, every step was receipted, and the finished record was checked against the rule stated before the run.
  • The work itself: 41 steps, each with its own receipt. The steps include a planning pass, a do, check, and fix loop that ran repeatedly, three live price lookups, and a source reachability check that passed. The record reports 7 sources resolved and 41 step receipts checked.
  • The recipient: one named human recipient, addressed by the wrap, not a vague "to whom it may concern."
  • The next steps, written on the record: the recipient opens the handover and checks any receipt himself, and states his orchestrator rule before his next run so drift is flagged on the step, not found on the bill.
  • The status: the handover was picked up by the receiving side. The client decision field sits on the same record and reads pending, which is exactly the honest state before a client signs off. Acceptance, when it happens, is recorded on this same record.

The record is also honest about its limits, and it prints them on the page: it does not prove an outside timestamp, it does not prove issuer identity, and it does not prove result correctness. A receipt proves the saved result was not changed after the fact. It never proves the result was right. That distinction is the whole product, and it is why a skeptical client can use the record instead of arguing with it.

The flow, step by step

  1. Do the work inside one job. Every tool call for the delivery happens inside a single job, so every call produces a receipt that belongs to that job. Work you run off to the side, outside the job, cannot be added later. The wrap refuses receipts that are not job owned, and that refusal is a feature: it stops you, and anyone else, from padding a delivery record after the fact.
  2. Wrap the job into a handover. A wrap requires four things: a summary, a goal, a recipient, and next steps. Leave one out and the wrap does not complete. When it completes, it binds the job's receipts into the handover, so the link you send carries the work and its proof together.
  3. Send the client one link. The client opens it and sees every step with its receipt, in order, with the summary and next steps at the top. There is nothing to install and no account needed to look. Pickup is always free.
  4. The client checks, then decides. The record's receipts were recomputed in the check above, and the same recompute is open to anyone holding the link. If the saved bytes still match the receipt, the result the client is looking at is the result the job produced. The client's sign off is then recorded on the same record they checked, so the approval and the evidence never live in two different places.

Now the invoice conversation changes shape. The question is no longer "prove it." The client has already run the check, or can run it while you are on the call. What is left to discuss is the work itself: scope, quality, and what happens next. Those are conversations that end in payment.

What it costs to try this

The free tier covers a first real delivery: every tool is free once a day, plus 3 continue calls shared across tools, and one free wrap up handover per day. If you deliver client work daily and hit the limits, the Day Pass is $0.99 for unlimited use for 24 hours, the Zambo Pass is $29 for 30 days, or $19 for 30 days for $ZAMBO holders, and the Lifetime Pass is $250. There is no annual tier.

This is not a small pilot system. The public stats endpoint at https://zambo.dev/api/stats, read on November 18, 2026 at 04:38 EST, reported 146,832 public receipts, with 865 of them created that day alone. The handover above is one of them, and your client's delivery can be the next one.

Send the proof before the invoice. Let the client check the work instead of trusting the summary, and let the sign off land on the same record as the receipts. That is how you prove AI work to a client before invoicing, in a way that survives the client checking it.

Add Zambo to your agent: https://zambo.dev/add/ (every tool free once a day, plus 3 continue calls, one free wrap-up handover per day, and pickup is always free). Hit the limit mid-delivery? Your work is saved: Day Pass, $0.99 for 24 hours.

Top comments (0)