The audit is an inventory, not a memory test
A Q4 audit fails when it starts from chat histories and screenshots. Chats get deleted, sessions blur together, and nobody can say in December what an agent did in March. Start from the record instead. If the year's agent work ran through Zambo, every call already carries its own receipt, and the audit becomes five passes over records that exist:
- Inventory the projects. List every project, client, and workflow the agents touched this year. Each workflow is one job with one record, so the inventory is a list of records, not a list of recollections.
- Find the missing records. For each project, ask which steps have receipts and which ran outside the connection. The gaps are findings in themselves: work you cannot show is work you cannot audit, and naming it now beats discovering it during a dispute.
- Check the approvals. Pull every handover and its acceptance state. Work that was accepted, work still pending, and work that was flagged and never resolved are three different year end stories, and the record separates them cleanly.
- Total the spend. Add up what the agents spent, where the caps held, and where a run hit its limit. Rejected calls mint receipts too, so the spend pass shows the attempts the cap stopped, not only the charges that landed. The routine version of this pass is the team AI usage audit.
- Close or carry every open handover. Anything still pending on December 31 gets a decision: accepted, flagged with a reason, or carried into January on the record, where the next session picks it up with full context.
Sample the receipts, do not reread the year
You do not need to reread twelve months of agent output. An audit samples. Pick the highest value projects, the costliest runs, and every flagged step, then recheck their receipts at zambo.dev/verify with no account. The check is simple: an agent receipt is a record of one ai agent tool call that anyone can check: it keeps the exact saved bytes and their sha-256 digest, and it proves the result was not changed since it was saved, not that it was right. A sample that rechecks clean tells you the record held. A sample that fails tells you exactly which project needs the deep look. The general method is in the AI work audit guide.
What the finished audit hands over
- A project index. Every project, its workflow record, and its final state, so anyone reviewing the year starts from the same list you audited.
- An exceptions list. Missing records, unresolved flags, spend limit hits, and rejected calls, each tied to its receipt, or to the named gap where no receipt exists.
- Client summaries that write themselves. Each project's handover already carries the goal, the finished steps, and the acceptance state. The year end summary is the handover, exported, not a report you reconstruct in January.
- A January starting state. Open handovers carry forward as living records. The new year starts from checked state, not from a folder of exports nobody trusts.
Zambo covers the calls routed through it: the audit is complete for the work that ran on the connection, and the gaps list shows you exactly what ran outside it.
Set up next year so the audit takes an afternoon
- Route the agents through the one endpoint. Connect each AI to https://zambo.dev/mcp so every call mints its receipt as the work happens. Start at zambo.dev/add: 20 free calls per day, no install.
- One workflow per project. A workflow per job gives the audit its index for free: one project, one record, one acceptance state.
- Review monthly, not yearly. The weekly version of this audit is the weekly agent review. Twelve small reviews make the Q4 pass a formality.
- Close handovers as you go. An Accept or a Flag recorded in the week the work landed is worth more than a reconstruction a year later.
Run next year's work on the record from the first call: add Zambo to your AI. Start with 20 free calls per day; the Day Pass is $0.99 USDC on Base for 24 hours of full use, and the Zambo Pass is $29 for 30 days, one payment.
Frequently asked questions
How do I do a Q4 audit of everything my AI agents did this year?
Work in five passes over the record. Inventory every project and workflow the agents touched, find the steps that have no receipts, check every handover acceptance state, total the spend including the attempts the caps stopped, and close or carry every open handover before December 31. If the work ran through Zambo, each pass reads records that already exist instead of reconstructing the year from chats.
What if some of the year's agent work has no receipts?
List it as a gap. Work that ran outside the recording connection cannot be rechecked after the fact, and the honest audit says so, project by project. The gaps list is still valuable: it shows exactly which work you can stand behind, and it tells you what to route through the connection next year.
How much of the year's work do I need to recheck by hand?
A sample, not the whole year. Recheck the highest value projects, the costliest runs, and every flagged step at zambo.dev/verify. Each receipt recomputes from its saved bytes, so a clean sample shows the record held, and a failed check points at the one project that needs the deep look.
What should the finished Q4 audit contain?
Three things: a project index with the final state of every workflow, an exceptions list covering missing records, unresolved flags, and spend limit hits, and a January starting state where every open handover is either accepted, flagged with a reason, or carried forward on the record.
What is an agent receipt?
An agent receipt is a record of one ai agent tool call that anyone can check: it keeps the exact saved bytes and their sha-256 digest, and it proves the result was not changed since it was saved, not that it was right.