A transfer record that travels with machine work when it moves between agents, tools, queues, or human reviewers.
// AGENT OPS · GTM CONTROL PLANES · June 6, 2026
Machine Workers Need Handoff Receipts Before Agent Work Crosses Systems
If work changes hands without a receipt, the next agent is operating on trust nobody earned.
Autonomous systems rarely fail in one dramatic action. They fail when small assumptions survive multiple handoffs and become operational truth.
Operators can reconstruct who knew what, what changed, and why the next worker was allowed to continue.
Agent work is becoming a relay, not a solo task.
Early automation was easy to inspect because one workflow owned the whole job. A script pulled data, transformed it, and produced an output. Machine workers do not stay that tidy. One worker researches. Another classifies. A third drafts. A fourth writes into a system. A fifth monitors the result.
That is where governance breaks. The riskiest moment is not always the original decision. It is the transfer. When work crosses from one context to another, the receiving worker sees an artifact but not the full shape of the judgment behind it.
A handoff receipt fixes that. It is not a meeting note. It is a control primitive for machine labor: the minimum evidence packet required before autonomous work can be accepted by the next actor.
The handoff is where assumptions get laundered.
Imagine a GTM control plane where an account-research worker marks a company as “active expansion potential.” The signal is based on hiring, product usage, and a funding event. A scoring worker receives the label and raises priority. A sequencing worker receives the score and drafts aggressive outreach. A CRM worker proposes a next-step field update.
By the end of the chain, nobody can tell whether “expansion potential” meant strong evidence, weak inference, stale context, or a human-approved account strategy. The label has been washed through enough systems that it feels official.
The problem is not that agents collaborate. The problem is that the collaboration has no boundary artifact. Without receipts, every downstream worker inherits confidence without inheriting proof.
What belongs in the receipt
Producer
The worker, workflow, version, tool set, and policy lane that created the transferred artifact.
Evidence
The sources used, freshness window, citations, conflicting data, and any inputs that were ignored or unavailable.
Decision state
Separate facts, inferences, recommendations, approvals, and system writes so downstream agents do not confuse them.
Open risk
Known caveats, low-confidence fields, privacy constraints, account sensitivity, and reasons the next step may need human review.
Acceptance
Who or what accepted the handoff, when it happened, and what actions the receiver is allowed to take next.
Grounded example
The campaign agent that inherited a bad premise.
A revenue team uses one agent to summarize customer interviews and another to draft reactivation campaigns. The summary agent flags “pricing concern” from a single old call. The campaign agent treats that tag as current buyer intent and generates discount-led outreach.
A handoff receipt would carry the source date, confidence, sample size, and unresolved caveat: one interview, eight months old, not confirmed by usage or renewal data. The downstream worker could still draft, but it would be forced into a safer lane: ask for confirmation, do not offer a discount, route for CSM review.
Receipts make multi-agent systems manageable.
Operators do not need to watch every token. They need reliable boundary checks. Handoff receipts let teams audit the points where work changes authority level: research to recommendation, recommendation to write action, draft to customer communication, exception to reusable policy.
They also make rollback possible. If a bad CRM update, support reply, or renewal recommendation appears, the team can trace the chain back to the handoff where the premise changed. That is how you fix the operating system instead of blaming the final agent in the chain.
Four handoffs that deserve receipts first
When raw research becomes a score, label, rank, segment, or recommendation.
When a proposal becomes a CRM write, workflow trigger, notification, sequence, or customer-facing draft.
When a reviewer approves, edits, rejects, or creates an exception that another worker may reuse.
When one autonomous worker hands partial work to another without a human sitting between them.
Multi-agent leverage is not more agents. It is clean custody of work as authority changes hands.
Operator audit
Where missing receipts show up
Downstream workers treat a tag, summary, or score as verified truth because the original confidence vanished.
Humans approve a final artifact without seeing the source quality or unresolved caveats that shaped it.
Teams know the output was wrong but cannot locate the boundary where the bad assumption entered the chain.
A one-off human approval travels to future workers without scope, expiry, or reuse permission.
Concrete takeaway for this week
Add handoff receipts to one agent chain before adding another worker to it.
- Artifact: what exactly is being transferred?
- Proof: what evidence and confidence travel with it?
- Delta: what changed since the last worker touched it?
- Risk: what caveat must the next worker preserve?
- Authority: what is the receiver allowed to do next?
The blunt rule: no receipt, no cross-system handoff. If your agents are already passing work without one, your company is scaling undocumented assumptions.