---
title: The week between Christmas and New Year
metaTitle: Holiday and out-of-hours email coverage with an AI agent
description: Coverage gaps are predictable, and so is what arrives during them. What an agent should do when nobody is available, what it must not promise, and how to hand back a backlog.
date: 2026-08-06
author: MCPmailer
tags: Operations
---

Every business has weeks where the inbox keeps filling and nobody is reading it. The usual answer is an autoresponder saying the office is closed, which tells the sender nothing they wanted to know and guarantees they write again.

![A full inbox and an empty office](/blog/holiday-coverage-email-agents/hero.webp "The mail does not stop. The staffing does.")

An agent can do better, but only if you decide in advance what "better" means, because the failure mode here is worse than silence.

## What arrives when nobody is there

The mix shifts during a shutdown, in ways that matter for how you configure things.

Routine questions continue at roughly their usual rate, and they are the easiest win: order status, opening hours, whether you are open at all, how to do something self-service.

Genuinely urgent things become more urgent, because the normal escalation path is closed. A production issue during a shutdown needs a person, and the agent's job is to reach one rather than to help.

And a third category grows: senders who know nobody is reading and write anyway, to have it in the queue for January. These need acknowledgement, not answers.

## The rule that prevents the bad outcome

An agent covering a gap must never make a commitment on behalf of an absent person.

"Someone will look at this tomorrow" is a promise nobody will keep on 27 December, and it is worse than saying nothing, because the sender now waits rather than escalating through another channel. Same for approvals, exceptions, and anything with a price attached.

What it can do honestly: answer what is factual, state clearly when the team is back, give the route that does work if something genuinely cannot wait, and confirm the message is in the queue.

| Message | Holiday behaviour |
| --- | --- |
| Factual question, answerable from data | Answer normally |
| Needs a decision | Acknowledge, state the return date, do not promise |
| Time-critical | Route to the on-call path, and say that it has been |
| Anything ambiguous | Acknowledge with the return date, queue for review |

## Say the date, not "shortly"

The single highest-value piece of copy in a holiday agent is a specific date.

"We are back on 5 January" lets somebody plan. "We will respond as soon as possible" does not, and it generates the follow-up message you were trying to avoid. Every acknowledgement should carry the return date, and it should be a variable rather than something written into a prompt someone forgets to update.

The same applies to out-of-hours coverage outside holidays: "before 10:00 tomorrow" beats "shortly", and it is checkable.

![An acknowledgement with a date, against one without](/blog/holiday-coverage-email-agents/dates.webp "A date is planning information. Shortly is not.")

## Tighten, do not loosen

The instinct when nobody is watching is to give the agent more latitude so it can cover more. Do the opposite.

Reduce what it may act on, lower escalation thresholds so borderline cases queue rather than get handled, and keep the send limits where they are, per [what happens when forty messages arrive at once](/blog/agent-email-concurrency-bursts). A backlog of correctly-queued messages is a morning's work in January. A week of confidently wrong answers is a quarter's problem.

The reason is straightforward: the safety net during normal operation is a person noticing something odd within a few hours. During a shutdown that net is gone, so the automated caution has to increase to compensate.

## Somebody still has to check

Fully unattended for two weeks is not coverage, it is a gamble. Even a light touch changes the risk profile completely.

Ten minutes a day, or every second day, reading the escalation queue and skimming what was answered. Whoever is nominally on call can do it from a phone, and it exists to catch the thing nobody anticipated rather than to review work.

Configure the escalation route to reach that person directly rather than landing in a shared inbox nobody is opening, which is the most common way holiday escalation quietly fails, per [what to monitor when an agent is answering your mail](/blog/monitoring-email-agents-in-production).

## The handback

The first morning back is where value is either realised or lost. Two things make it work.

The queue should be ordered by what needs a decision rather than by arrival time, since a fortnight of chronological backlog is exactly what makes January miserable.

And each queued thread should carry the agent's note on why it was held: what was asked, what it could not decide, and what it already told the sender. That turns a fortnight of reading into a morning of deciding, per [the weekly half hour that keeps an agent honest](/blog/weekly-review-email-agents).

## Questions

### Can an AI agent cover email during a holiday shutdown?

Yes, for factual questions, acknowledgements, and routing anything urgent. It should not make commitments on behalf of people who are away.

### What should a holiday acknowledgement say?

The specific return date, what the sender can do if something cannot wait, and confirmation that the message is queued. Never "as soon as possible".

### Should I give the agent more autonomy while the team is away?

Less. The safety net of someone noticing an odd reply within hours is gone, so lower thresholds and let borderline cases queue instead.

### Does someone still need to check in?

Yes. Ten minutes a day on the escalation queue, reachable directly rather than through a shared inbox nobody is opening.

### How should the backlog be handed back?

Ordered by what needs a decision rather than by arrival time, with the agent's note on each held thread explaining what was asked and what the sender was already told.

### What about ordinary out-of-hours coverage?

The same design, with a shorter horizon. A specific time beats a vague one, and the on-call route should be real rather than nominal.

## Related

- [The weekly half hour that keeps an agent honest](/blog/weekly-review-email-agents)
- [What to monitor when an agent is answering your mail](/blog/monitoring-email-agents-in-production)
- [When an email agent is the wrong answer](/blog/when-not-to-use-an-email-agent)
- [Designing approval that a human will still read](/blog/human-in-the-loop-email-approval)