Governed Agent Labor Needs Custody Chains Before Agents Handoff Customer Work
Do not let autonomous handoffs erase accountability just because the next worker picked up the ticket.
WhatCustody chain
A governed handoff record that preserves owner, intent, evidence, authority, and acceptance when customer work moves between workers.
WhyHandoffs lose truth
Agents can move work instantly, but customers experience the gaps when promises, exceptions, and risk notes do not travel with the task.
ConsequenceSafe delegation
Operators can let machine teams coordinate without letting responsibility dissolve across invisible internal transfers.
The dangerous part of agent work is often the handoff.
Most teams worry about whether an individual agent can complete a task. That is the easy part to inspect. The harder failure appears when two or more workers share one customer outcome.
A prospect asks for a security exception. A support ticket includes an angry renewal threat. A customer success note says, "do not mention pricing until procurement replies." Each item can be routed, summarized, tagged, and processed. The problem is not motion. The problem is custody.
Governed agent labor needs custody chains because the company must know what traveled with the work, what got dropped, and who accepted the risk before the next machine worker acted.
Four things must survive every handoff.
01Intent
The receiving worker needs the business purpose, not just the task label. "Reply to customer" is not the same as "protect renewal while legal reviews an exception."
02Evidence
Call notes, CRM history, contract terms, product usage, support severity, and prior promises must transfer as cited inputs, not vague summaries.
03Authority
The next worker must know whether it can draft, send, edit data, offer terms, escalate, or only prepare a recommendation for review.
04Owner
Someone remains accountable for the customer outcome. Handoffs should change execution state, not make responsibility anonymous.
Example: the expansion deal that becomes three disconnected tasks.
A growth agent identifies a high-usage account and opens an expansion motion. It passes the opportunity to a sales worker. The sales worker drafts outreach, but the customer has an unresolved support issue. A support agent then summarizes the ticket and sends it to customer success. Success asks RevOps to update health status.
Every individual action looks reasonable. The workflow is fast. The handoffs are automated. But nobody can answer the basic custody questions: which worker knew about the unresolved incident, who was allowed to contact the customer, which promise had already been made, and who decided expansion pressure was still appropriate?
Without a custody chain, the company creates a customer experience that feels coordinated internally and careless externally.
The custody ledger
Work item
Name the customer, account, ticket, opportunity, record, or campaign object being transferred.
Sending worker
Record which agent, automation, or human last held the work and what action it completed or attempted.
Evidence packet
Attach cited facts, source links, timestamps, confidence, missing context, and known contradictions.
Authority scope
State what the receiving worker may do: observe, draft, enrich, update internal fields, message a customer, or request approval.
Acceptance mark
The receiver must explicitly accept custody or reject it with a blocker. Silent handoff is not custody.
Custody chains are not bureaucracy. They are memory with teeth.
Fast machine work makes it tempting to treat every transfer as internal plumbing. That is fine for low-risk enrichment. It is reckless for customer-facing work, revenue truth, legal commitments, data changes, and escalations.
A custody chain does not slow the whole system. It slows ambiguity. It makes the receiving worker inherit the actual state of the work instead of a flattened instruction that hides why the task matters.
Once custody is explicit, operators can review handoff quality, find repeated context loss, detect unauthorized action, and prove which worker made which customer-impacting decision.
Autonomous teams do not need more motion. They need accountable transfer.
Operator checks
Three signs your agents need custody chains
01Customers repeat themselves
Agents keep asking for context the customer or another internal worker already provided.
02Promises become folklore
Exceptions, commitments, and constraints are mentioned in summaries but cannot be traced back to source evidence.
03Owners disappear
When something goes wrong, the audit trail shows many actions but no accepted owner for the customer outcome.
Install one custody chain this week.
ChooseOne handoff
Start with support to success, growth to sales, sales to RevOps, or agent to human review.
PackageEvidence
Require source links, confidence, missing facts, and prior promises to travel with the task.
LimitAuthority
Define what the receiver can do before approval and what must escalate.
MarkAcceptance
No worker acts until custody is accepted or rejected with a visible reason.
Concrete takeaway for this week
Put a custody chain on one recurring customer-facing agent handoff.
Name the work: identify the account, ticket, opportunity, or record being transferred.
Preserve intent: include the business goal and customer risk behind the task.
Move evidence: attach source material instead of trusting compressed summaries.
Constrain action: state what the receiver may do before human or policy approval.
Require acceptance: make the next worker explicitly accept custody before acting.
The blunt rule: if agents hand off customer work, custody must move with the task or the company is delegating into fog.