Let software help. Keep the say.
Someone hands a task to software that will decide for itself how to do it. Everything that matters happens in the parts nobody demonstrates: what it was allowed to touch, what it did, and how it is stopped.
Help plan next week.
Choose what the assistant may do.
A plan for next week
A draft is ready for you.
Interactive example · permissions and schedules are illustrative
The situation
Delegated software is useful and different from a script in a way that matters. A script does what it was told. An agent chooses an approach, which means the interesting question is no longer whether it works but whether the organization can still hold anything accountable afterwards.
How this usually goes
The agent runs with a human's credentials, so the audit trail says a person did it.
Permissions are broad because narrowing them was fiddly and the task was urgent.
Success is whatever the agent reported, checked by nobody.
Stopping it means finding out where it runs, which is a conversation rather than a control.
What happens here
1. The task is scoped before it starts
What is to be achieved, what may be touched, what must not be, and what would count as done. A task that cannot be scoped is a task that is not ready to be delegated.
2. Authority is granted narrowly and expires
The agent holds its own identity and is given the rights this task requires, for as long as it requires them. Standing broad access is treated as the failure mode it is.
3. Work happens in an isolated space
Changes are made somewhere they can be examined and discarded, not directly in the thing that people depend on.
4. The checks were defined in advance
Verification is decided before the work, so completion is measured against a standard the agent did not choose for itself.
5. Acceptance is a separate act
Nothing lands because the agent believed it was finished. A reviewer, human or a differently-scoped validator, accepts or rejects, and that decision is recorded alongside the work.
6. Revocation is designed
What stops, what is already in flight, what must be reconciled, and how the organization can see that it took effect. This is built before delegation is used, not after an incident.
What is on the record afterwards
What was asked, which authority permitted it, what was done, what was checked, and who accepted it. That is the difference between automation an organization governs and automation that quietly becomes the only thing that understands its own systems.