If you deliver client projects built with AI coding agents, the handover is where your margin and your reputation both live. This page is the delivery standard: what changes hands, in what form, and how to make the last step of every project the moment the client decides to hire you again.
Clients cannot evaluate your code. They evaluate what delivery feels like: whether they can get into everything, whether things work when your team is not on the call, and whether the project survives the relationship ending. AI-assisted building sharpens this, because your throughput went up while the client’s ability to assess the work stayed exactly where it was. The agencies winning repeat work in this market are the ones who close that gap deliberately, with a handover that reads like proof instead of a farewell.
The other commercial reality: a defined handover is a boundary. Without one, delivery blurs into indefinite free support, because the client has no way of knowing what was finished. A clean, evidenced close is how what was delivered stops being a matter of memory.
The full item-by-item list, with a free template, lives in the AI project handover checklist; it covers the repository, the executed setup, the knowledge and the credentials. For an agency the point is not the list, it is that the list is the same on every project. A repeatable standard means a junior can run the delivery, the client sees the same professional close every engagement, and nothing depends on which developer happened to remember what. Write your version down once, make it part of the definition of done, and treat a project as deliverable only when its handover artefacts exist.
The strongest close we know of has three parts. Deliver the repository and access properly, with the client logging into every service themselves on the call. Hand over a project passport: the evidence document a free tool generates from the repository and your agents’ session history, with setup executed in a clean sandbox and every statement labelled by how it was established. Then, and this is the part clients repeat to other founders, show them the check. One command on their side compares the delivery against the passport and re-runs the recorded checks:
You are handing the client the means to audit you, which is exactly why it works. Trust me is what every agency says. Check me is a differentiator, and it costs you a ten-minute run at the end of a project. The client-side explanation to send with the link is the recipients page, written so a non-technical stakeholder can read the labels without a translator. When a client or their investor wants the proof to come from neither side, a Mantyl Verified run signs the re-execution on independent infrastructure for a one-off £19 to £49, which is the kind of line item that looks very good on a delivery invoice.
Handover work is real work; the mistake is doing it as unpaid goodwill at the end. Because the evidence half is automated, a professional handover costs you minutes of tooling plus the human half of the checklist, and it belongs in the quote as a named deliverable: clients pay happily for the thing that protects them. Per-delivery verification also matches how agency economics actually run, which is why Mantyl charges per project rather than per seat. The tiers, including where the team plan with white-label passports is heading, are on the pricing page.
End every delivery by stating the support arrangement in writing, even when the answer is generous, even when it is none. The passport helps here too: with the known gaps and unknowns listed at delivery, later conversations start from the shared record instead of from competing recollections. That is the quiet, compounding value of evidence-first delivery: it makes the end of every project the least stressful part, for both sides, which is precisely the memory you want a client left holding. The broader case for treating handover as an engineering discipline is in the AI code handover guide.