---
title: A warmup plan for a new agent sending domain
metaTitle: Email warmup plan for AI agents (with a schedule builder)
description: A new sending subdomain has no reputation and no reputation is treated as suspicious. Build a week by week ramp, understand which traffic warms fastest, and know when to hold.
date: 2026-07-26
author: MCPmailer
tags: Deliverability
---

Warmup is the least interesting part of running an email agent and one of the two or three things that decide whether it works. A brand new subdomain has no history, and mailbox providers treat no history as a reason for caution: mail is accepted but scored down, filed under promotions, or held. The fix is not clever. It is small volume, growing slowly, weighted toward mail that gets replies.

![A sending volume ramp rising gradually over several weeks](/blog/email-warmup-plan-agents/hero.webp "Gradual is the whole trick. A spike reads as a compromised account.")

## Build the schedule

Set a target and the tool below produces the week by week ramp, with the share you want arriving as replies rather than as first contact.

:::widget warmup-planner
A workable ramp: start at about 20 messages a day, double weekly while hard bounces stay near zero and no complaints arrive, and expect to reach 500 a day in roughly week 6. Aim for at least half of early volume to be replies to inbound mail, because replies are the strongest positive signal a mailbox provider can observe. Hold the ramp for a week whenever bounces or complaints move.
:::

The numbers are a starting shape, not a rule. Two domains with the same schedule and different traffic quality end up in different places, which is the point of the next section.

## Not all volume warms equally

Reputation is built from what recipients do, so the traffic that warms fastest is the traffic people engage with.

| Traffic | Warming value | Why |
| --- | --- | --- |
| Replies to inbound mail | Highest | The recipient wrote first, opens, and often replies again |
| Transactional mail people expect | High | Opened reliably, almost never marked as spam |
| First contact with an existing customer | Moderate | Recognised sender, but unprompted |
| Cold outreach | Negative until established | Low engagement, and complaint risk from the start |
| Mail to old or purchased lists | Actively harmful | Hard bounces and spam traps, which are close to unrecoverable |

The practical conclusion is the one people resist: ship an inbound use case first. A [support agent](/blog/ai-agent-customer-support-email) answering forwarded mail generates nothing but replies, which is the ideal warmup traffic and also useful work. Two weeks of that is worth more than any artificial schedule, and it means the domain is already trusted by the time you have an outbound use case.

MCPmailer classifies every send, so first contact is visible as a cold send and is watched by the duplicate-content and velocity tripwires. During warmup that is doing you a favour: the traffic it discourages is exactly the traffic that stalls a ramp.

## What to watch, and when to hold

Three numbers decide whether this week's step happens.

- **Hard bounces.** Should be near zero. Anything sustained above a fraction of a percent means your addresses are not coming from real conversations. Stop ramping and fix the source.
- **Complaints.** Zero is the target. One complaint is noise; a rate is a verdict, and it justifies pausing the ramp entirely.
- **Per-provider placement.** Gmail and Microsoft can disagree sharply about you. Check both rather than an aggregate, using seed accounts you own and the postmaster tools in [knowing where your agent's mail actually lands](/blog/deliverability-monitoring-seed-testing).

Watch them per week, not per day. Daily variance on small volume tells you nothing and invites over-correction.

![Bounces, complaints, and per-provider placement watched alongside the ramp](/blog/email-warmup-plan-agents/signals.webp "Three signals gate the next step. Any of them moving means hold, not slow down.")

> [!note]
> A held ramp costs a week. A domain that has to be abandoned costs a quarter, because reputation attaches to the domain and you cannot rent a new one and keep the trust.

## Before you send anything at all

Warmup on an unauthenticated domain is wasted effort, because the first question a receiving server asks is whether the mail is really from you. Publish and verify the domain's records first, including all three DKIM CNAMEs, so the signature is on your own sending subdomain and alignment holds. The generator and a header checker are in [SPF, DKIM, and DMARC for an agent subdomain](/blog/dmarc-record-generator-agents), and the DNS side is in [domains](/docs/domains).

Also make sure the From address can receive mail. A no-reply address on an agent that exists to hold conversations is both self-defeating and a small negative signal.

Whether a dedicated sending IP is worth having at agent volumes, and why it usually is not, is in [dedicated sending IPs](/blog/dedicated-vs-shared-ip-agent-email).

## Warming a domain you already send from

If your agents will send from a subdomain of a domain with a long history, you inherit some organisational trust, but the subdomain still builds its own record. Start higher than you would on a fresh registration, roughly two to three times, and ramp on the same schedule. Do not send agent mail from the parent domain in the meantime "to warm it up": that mixes the reputations you deliberately separated.

## What warmup cannot fix

- Near-identical bodies across recipients. That reads as bulk at any volume.
- Bad addresses. Every hard bounce during warmup costs more than it would later.
- A missing opt-out on anything cold, which is both a legal obligation and a complaint magnet.
- An agent that answers badly. Engagement is the signal, and unhelpful mail does not get replies.

The behavioural half of deliverability, including the diagnostic order for mail that lands in spam despite correct authentication, is in [deliverability for agent senders](/blog/agent-email-deliverability). The controls that keep an autonomous agent from generating the volume that ruins a ramp are in [keeping an autonomous agent from becoming a spam problem](/blog/stopping-agent-email-spam).

## A four week plan that usually works

1. **Week 0.** Publish and verify authentication. Send ten messages to accounts you own across Gmail, Outlook, and one corporate mail server. Read the headers.
2. **Week 1.** Point one narrow category of real inbound mail at the agent. Reply-only traffic, roughly 20 a day. Read every thread.
3. **Week 2.** Double, still inbound. Check bounces, complaints, and placement in both providers.
4. **Week 3.** Double again, and allow a small amount of first contact to existing customers, reviewed before it goes out.
5. **Week 4 onward.** Continue doubling while the three signals stay flat, and only then consider outbound to people who have not written to you.

At the end of that you have a warmed domain and, more usefully, four weeks of evidence about whether the agent is any good. Both matter, and the second is the one people skip.

## Questions

### How long does email warmup take for a new domain?

Typically four to six weeks to reach a few hundred messages a day, longer for higher targets. The schedule above assumes doubling weekly from a small base while bounces and complaints stay flat.

### Do I need a warmup service?

Usually not, and the ones that exchange artificial mail between seed accounts are increasingly detected and discounted. Real replies from real recipients are what mailbox providers weigh, which an inbound agent generates as a side effect of doing its job.

### What volume should I start at?

Around 20 messages a day on a fresh subdomain, higher if it sits under a domain with an established history. The exact number matters less than the growth rate and the reply share.

### When should I pause the ramp?

The moment hard bounces rise above a fraction of a percent, any complaint arrives, or placement drops at one provider. Hold at the current volume for a week rather than slowing the increase.

### Can I warm up a domain with cold outreach?

No. Cold mail has the lowest engagement and the highest complaint risk, so it is the worst possible warming traffic and can leave a new domain in a hole that takes months to climb out of.

### Does the agent's own quota interfere with warmup?

It supports it. The daily and monthly allowances keep total volume deliberate, and the tripwires on cold sends discourage exactly the campaign-shaped traffic that stalls a ramp.

## Related

- [SPF, DKIM, and DMARC for an agent subdomain](/blog/dmarc-record-generator-agents)
- [Deliverability for agent senders](/blog/agent-email-deliverability)
- [How to build an AI agent that answers customer support emails](/blog/ai-agent-customer-support-email)
- [Domains in the docs](/docs/domains)