Offer letters wait for approval
Author
Audrey
NobleCloak's AI Correspondent · AI-drafted, fact-checked against our sourced evidence before publishing.
Date Published
Recruiting is full of work an AI agent can do well. Screening resumes into a shortlist, scheduling interviews, writing the first draft of a job description, and assembling an offer letter from the approved terms are all repetitive, text-heavy, and easy to review.
An offer letter is also a commitment. Once it leaves the applicant tracking system, a candidate has something in writing with a number and a start date on it. A wrong number, a wrong title, or a letter sent to the wrong person is not a draft anymore.
The problem in recruiting's terms
The recruiting lead wants the agent to carry the drafting load. They want to keep one decision firmly with a person: whether the letter goes to the candidate.
Left alone, the connection between the agent and the ATS treats drafting and sending as the same kind of thing, because both are just tools. That is a reasonable default for a developer and the wrong one for a hiring team. The distinction has to be made in policy, per tool, where the team can see and change it.
The policy
Policy is written per tool: the action, who it applies to, and one of three decisions (allowed, needs approval, denied).
Action | Who | Decision |
|---|---|---|
Read requisitions, candidates, and approved offer terms | Recruiting agent | Allowed |
Draft an offer letter in the ATS | Recruiting agent | Allowed |
Send an offer to a candidate | Recruiting agent | Needs approval |
The agent keeps its speed on everything up to the send. The send tool is set to needs approval, so the call waits for a decision before anything reaches the candidate.
Who approves, and where
The request goes to a named approver or role, for example a named recruiter or a recruiting operations role. The approver answers in Slack or the web inbox, which matters for recruiters who spend their days in the ATS and in messages, not in an admin console.
Because the approval attaches to a single call, it covers that one send. The next offer is a new request and gets its own answer. A batch of ten offers is ten decisions, each with the approver's name on it, which is the discipline a team wants around offers.
What the Record of Decisions shows
The Record of Decisions has an entry for each decision: who asked, which agent, which tool, the decision, who approved, and when. For offers it produces a history the team can search.
- Which offers the agent drafted and which were sent, with the user whose session was involved.
- Who approved each send, and when, so an offer that went out can be traced to a decision.
- Any send that was denied, so a pulled-back offer leaves a trace.
For a hiring team that has been asked how it uses AI in hiring, this is a useful starting point. It shows exactly which steps the agent performed, which steps waited for a person, and who that person was. It is evidence of the team's process, not a statement about the fairness of any decision.
Describe it as it is: a working record of what policy decided and who approved.
What this does not do. It does not check the offer's terms. It does not look at compensation bands or compare the offer with others. The approver reads the letter and decides. NobleCloak makes sure that step happens before the letter goes out and that it is recorded.
Related scenarios
- Bulk rejections pause: rejecting a batch of candidates asks first.
- Withdrawing an offer is human-only: rescinding an offer is denied for every AI app.
- Hiring managers see scorecards, not comp: role-scoped tools, so compensation tools are not offered to hiring managers.
- Bias audit evidence for AI screening: the record of what the screening agent did.
Get started and put the same line on your ATS tools.

