People and agents
Make responsibility clear before the decision
Good collaboration can still leave approval unclear. What decision-making practice teaches us about authority, evidence and a record people can rely on.
A discussion can move forward without a decision
People can agree that a proposal is possible without agreeing that the business should commit to it. One person confirms capacity, another estimates the cost and a third prepares the customer response. Each contribution is useful. Together, they can still leave the authority to proceed unresolved.
This becomes expensive when the next step creates an obligation. A message is sent, a purchase is made or an agent changes customer terms. The business then has to establish whether the people involved were advising, approving or simply keeping informed.
Our conclusion is that responsibility needs to be explicit at the point of commitment. Reconstructing the discussion afterward is useful for review, but it cannot provide the missing approval before the action.
Separate expertise from authority
Atlassian’s DACI framework makes a helpful distinction between the person driving a decision, the person approving it, the people contributing expertise and those who need to be informed. Its guidance also calls for recording the outcome so future teams can understand the reasoning.
The useful principle extends beyond a particular framework. A person can know whether a change is technically possible without owning its commercial consequences. A Finance estimate can inform a decision without approving the expenditure. Making these roles clear protects contributors from having helpful advice mistaken for authorization.
That clarity also helps the approver. The review can focus on a defined proposal, its consequences and the remaining questions. People spend less effort interpreting the tone of a thread or deciding whether silence meant consent.
Sources: [1]
Carry the conditions with the approval
An approval is often narrower than a simple yes. It may apply to a particular amount, customer, deadline or version of an agreement. If any of those details change, the business needs to know whether the earlier decision still applies.
A useful record therefore connects the approval to the request that was reviewed. It retains who decided, what they were allowed to decide, which evidence they considered and any conditions they attached. This lets the next person understand the boundary of the decision without reopening every conversation.
For a connected AI workflow, the same principle needs to be enforced before execution. A general instruction to help a customer does not authorize every concession. A request outside the agent’s authority should reach a qualified reviewer with enough context to decide.
Accountability should make delegation easier
NIST’s AI Risk Management Framework Playbook treats documented responsibilities, policies and oversight as part of governing AI risk. We take a practical lesson from that guidance: introducing agents should make responsibility easier to inspect, rather than distribute it across more systems and people.
Clear boundaries let routine work proceed within agreed authority. Exceptions can go to the right person, instead of making the owner of the business the approval queue for everything. That is the operational benefit of governance done well.
The record matters later, too. An auditor reviewing an action should be able to connect it to the applicable rule version and the responsible reviewer. They should also be able to see where the evidence stops. Approval establishes a decision; confirmation from the connected tool establishes whether the action happened.
This is how we think about accountability at Orcinta. It should help a business delegate confidently today and explain a consequential decision tomorrow.
Sources: [2]
People should know what they can decide, and which decisions need someone else.
Sources and further reading
- 1DACI decision-making framework
Atlassian Team Playbook
- 2