Why does delegation to AI need a framework?
Delegation to an AI employee needs a framework because the failure modes are the same as for any delegation: unclear scope, no checkpoint, too much trust too soon, and no review. Managers who delegate well to people already have the instincts; the framework makes them explicit for a worker that never pushes back and never asks for clarification unless designed to.
On ftes.ai the four parts map to mechanisms: the remit in Setup and Targets, the approval gate in Approvals, the autonomy ladder in Setup, and the review in Performance. The framework is what a manager does with them.
How do you write the remit?
Write the remit in three lists. What the Digital FTE owns: the function, the properties, the tools, the outputs it is accountable for, which become its Targets. What it escalates: the categories that always reach a person, regardless of confidence, and the triggers such as no source found. What it never does: the actions that are permanently gated, such as anything affecting a person's outcome.
Keep it short and specific. A remit that says handle support is not a remit; one that says resolve billing-question and how-to tickets from the knowledge base with a cited source, escalate refunds, account access, complaints, and distress, never issue credits, is.
How do you set the gate?
Set the gate by starting at Level 0 on the autonomy ladder, where every action is a proposal that waits for a named approver. Then decide who approves what: the function's manager for most actions, a specialist for specific categories such as a clinician for clinical content or the finance lead for adjustments.
Approve, edit, or send back with a reason. The reason is the important part, because send-back reasons are the data the remit is refined from and the ladder is climbed on.
- Start at Level 0: everything gated
- Name the approver per category
- Always give a reason on send-back
- Review send-back reasons weekly in the first month
How do you climb the ladder?
Climb the ladder one class of action at a time, from evidence. When a class such as exact-match reconciliations or interview scheduling shows a long run of approvals with no edits, promote that class to auto-approve. Leave every other class gated. If a promoted class produces a send-back, step it back and look at why.
Never promote a class on the always-escalate list, and never promote blanket. The ladder is per class precisely so that trust is granted where it has been earned and nowhere else.
How do you review?
Review on a cadence: weekly in the first month, then monthly. The Digital FTE writes a first-person monthly report in its Performance tab stating what it delivered against Targets, what it missed, what is unmeasured, and what it recommends. Read it alongside the send-back reasons, usage per role, and the approval rate.
Decide three things at each review: does the remit need changing, does any class deserve promotion or need stepping back, and are the targets right. Record the decisions. That record, over a year, is the history of how the delegation matured, and it is what a new manager inherits.
What does the framework look like in the first week?
In the first week the framework is four short acts. The remit is written in the first-run wizard, which pre-fills what the platform already knows and asks only for what is new to the role: the three lists of what it owns, escalates, and never does, and the targets it will be held to, proposed by the Digital FTE and approved by the manager. The gate is set by default, because every class of action starts at Level 0, and the manager names who approves which category so that routine items reach the team member closest to the work.
The ladder is not climbed in week one; the manager reads the morning brief daily, approves or sends back with reasons, and lets the record build. The review is scheduled: the send-back reasons weekly for the first month, the full review monthly from the FTE's own report. A manager who completes the four acts in week one has delegated properly, and the rest of the framework is the habit of reading what the FTE reports and deciding from it, which the workspace is built to make short.