A decision rights ledger is a living record of who can approve, block, delegate, escalate, or reverse agent work in each operating lane.
The right to say yes is not the same as being mentioned on a ticket or owning the next step in a queue.
// AI-NATIVE ORG DESIGN · GOVERNED AGENT LABOR · June 20, 2026
AI-Native Org Design Needs Decision Rights Ledgers Before Agents Reassign Accountability
If an agent can move responsibility without a ledger, the org chart becomes theater and accountability becomes fog.
Autonomous workers are good at completing paths. Without explicit rights, they infer ownership from activity, titles, queue position, or prior behavior.
Teams scale machine labor without letting responsibility leak from named humans into vague workflows and orphaned decisions.
Autonomous work breaks first at the boundary between speed and authority.
Most agent failures are described as execution problems. The worker sent the wrong message, updated the wrong field, or took the wrong next step. Underneath, the deeper issue is usually simpler: the agent did not know who had the right to decide.
In a human team, decision rights are carried through memory, politics, meetings, and tone. People know that finance can veto discount logic, legal owns risky promises, customer success can pause an expansion sequence, and the founder can override the normal route when a strategic account is involved.
Machine workers do not inherit that texture safely. They see tickets, roles, Slack activity, CRM ownership, prior approvals, and deadlines. Then they optimize the path. If the path is missing a decision rights ledger, the agent can quietly reassign accountability while everyone thinks it is merely moving work.
Four rights agents should never infer.
The right to block work needs explicit authority. Otherwise agents treat silence, delay, or missing feedback as soft approval.
Passing work to another person or worker changes load, context, and consequence. It should leave a receipt.
The right to break policy for a customer, segment, or deadline is a business decision, not an optimization shortcut.
Example: the expansion agent that accidentally moves accountability.
A growth agent notices an enterprise customer has expansion intent. It drafts a renewal motion, routes a discount exception to the account owner, and sees no response for two days. The customer success manager has been active in the account thread, so the agent asks them to approve the commercial language and moves the sequence forward.
Nothing looks dramatic. The customer gets a timely answer. The CRM updates. The sequence continues. But accountability has shifted. The CSM may understand relationship risk, not margin policy. The account owner may still be responsible for the number. Finance may never see the exception. If the promise later hurts the deal, the agent log says the work was approved, while the humans disagree about who had the right to approve it.
A decision rights ledger would stop the jump. It would show that the CSM can add customer context, the account owner can approve commercial posture, finance owns discount exceptions, and timeout after two days means escalate rather than substitute a nearby human.
What belongs in the decision rights ledger.
Decision type
The class of decision: approval, rejection, exception, customer promise, budget movement, forecast change, delegation, escalation, or rollback.
Named holder
The person, role, committee, or system that currently holds the right. Not a channel, not a vibe, not whoever replied last.
Scope boundary
The accounts, segments, spend levels, severity levels, countries, teams, or data classes where that right applies.
Evidence price
The proof the agent must attach before using or requesting the right: customer context, policy citation, revenue impact, risk note, or prior approval.
Expiry and fallback
When the right expires, what silence means, where escalation goes, and what the worker must do when holders disagree.
Decision rights are infrastructure, not bureaucracy.
Bad operators hear “rights ledger” and imagine paperwork. Good operators see a routing table for accountability.
Agents move fast because they can carry context across systems. That same speed makes informal authority dangerous. A machine worker can treat a comment as approval, a past exception as policy, or a late reviewer as permission to continue. The ledger converts hidden social rules into machine-readable operating law.
This is especially important in AI-native org design because the team will not stay stable. Agents will join, split, specialize, and supervise other agents. Humans will move from doing work to setting constraints. Without explicit decision rights, the org becomes fast but legally and operationally mushy.
The agent can do the work. The ledger says who owns the consequence.
Operator checks
Three signs your agents are already moving rights
Workers continue when owners do not reply, but nobody has defined whether timeout means approve, pause, escalate, or cancel.
The agent asks the most available human instead of the person with actual authority over money, risk, policy, or customer commitment.
The log says “approved” without naming the holder, decision type, scope, evidence, and expiry of that approval.
Install the first ledger before the next agent rollout.
Start with discounts, outbound offers, forecast edits, support exceptions, customer promises, or handoffs between sales and success.
Separate approval, rejection, delegation, escalation, exception, and rollback. Do not bundle them into “owner.”
Attach each right to segment, threshold, system, region, customer tier, deadline, and evidence requirement.
Make the worker record which right it used, who held it, what proof supported it, and when the decision expires.
Concrete takeaway for this week
Before giving an autonomous worker more operating range, build one decision rights ledger for one risky workflow.
- List the decision types: approve, reject, delegate, escalate, make an exception, promise a customer, or reverse work.
- Name the holder: the exact role or person allowed to make each decision.
- Define scope: where the right applies and where the agent must stop.
- Set silence behavior: pause, escalate, expire, or continue only under a named fallback.
- Require receipts: every agent decision should show right used, evidence, holder, timestamp, and expiry.
The blunt rule: agents may execute fast, but they should never invent who is accountable.