what is email warmup and why your AI agent needs it

You just gave your agent a fresh email address. It sends 300 outreach messages on day one. By day two, Gmail is routing every single one to spam. By day three, your domain reputation is so damaged that even your personal emails from the same domain start disappearing.

This is what happens when you skip email warmup. And it happens to AI agents far more often than it happens to humans, because humans don't usually try to send 300 emails from a brand-new address on their first day.

What email warmup actually is#

Email warmup is the process of gradually increasing your sending volume from a new email address or domain so that mailbox providers (Gmail, Outlook, Yahoo) learn to trust you. Think of it like a credit score for email. A new address has no history, no reputation, and no trust. Every major email provider treats unknown senders with suspicion by default.

When you warm up an address, you start by sending a small number of emails to real recipients who engage with them: opening, replying, moving them out of spam if they land there. Over days and weeks, you slowly increase the volume. The mailbox providers observe this pattern of low volume plus positive engagement and gradually assign your domain a better reputation score.

The typical warmup schedule looks something like this:

• Week 1: 5-10 emails per day
• Week 2: 20-30 emails per day
• Week 3: 50-75 emails per day
• Week 4: 100-150 emails per day

The exact numbers vary depending on who you ask. MailReach, Instantly, and other warmup tools all have slightly different ramp curves. But the principle is the same: start small, grow slowly, and make sure recipients are engaging with your messages.

Why mailbox providers care about this#

Gmail alone processes over 1.8 billion accounts and filters billions of messages daily. Their spam detection doesn't just look at the content of your email. It looks at your sending patterns, your domain age, your authentication records, and whether recipients interact with your messages or immediately delete them.

A brand-new domain sending hundreds of emails on day one looks identical to a spammer. There's no prior history to suggest otherwise. The provider has one data point (a sudden blast of email from an unknown sender) and one reasonable conclusion (this is probably spam).

Warmup gives providers the data they need to make a different conclusion. Consistent volume, positive engagement, proper authentication, and gradual growth all signal "legitimate sender." Without that signal, you're starting every conversation in the spam folder.

Where AI agents go wrong#

Human email users warm up their addresses naturally. You set up a new work email, send a few messages to colleagues, get replies, maybe send a newsletter to 50 people after a week. The warmup happens without you thinking about it.

AI agents don't work that way. An agent provisions an inbox and immediately starts executing whatever workflow it's been assigned. If that workflow involves sending emails, the agent will send as many as it can as fast as it can. There's no built-in concept of "take it slow for the first few weeks."

This creates three specific problems:

Volume spikes from cold addresses. An agent that sends 200 emails on its first day will get flagged. Not might get flagged. Will. The math doesn't work out any other way when your domain has zero reputation.

No engagement loop. Warmup works because recipients open and reply to your early messages. Most agent emails are transactional or outbound, meaning the engagement rate is naturally lower. If your first 50 emails get zero opens and zero replies, your reputation drops instead of building.

Multiple inboxes compounding the problem. Agents often create several inboxes for different purposes. Each one starts with zero reputation. If three inboxes on the same domain all start blasting simultaneously, the domain-level reputation tanks even faster than any individual address would.

The authentication foundation#

Warmup alone isn't enough if your DNS records aren't set up correctly. Before you send a single email, your domain needs three things:

1. SPF (Sender Policy Framework) tells receiving servers which IP addresses are allowed to send email on behalf of your domain.
2. DKIM (DomainKeys Identified Mail) attaches a cryptographic signature to each message, proving it hasn't been tampered with in transit.
3. DMARC (Domain-based Message Authentication, Reporting, and Conformance) tells receiving servers what to do when SPF or DKIM checks fail.

Without these records, your emails fail authentication checks before reputation even enters the picture. A lot of agent setups skip this step entirely because the agent doesn't know it needs to happen, and the developer assumed the email provider handled it automatically.

Some providers do handle it. If you're using a service like LobsterMail where the agent provisions inboxes on a shared domain that already has proper authentication, the DNS part is taken care of. But if you're pointing agents at custom domains, you need to set up these records yourself before warmup begins.

Automated warmup tools vs. manual warmup#

The warmup industry has grown significantly over the past two years. Services like MailReach, Instantly, and Warmy.io offer automated warmup where their network of real email accounts exchanges messages with your new address. They open your emails, reply to them, and mark them as "not spam" if they land in junk folders. This simulates the organic engagement that mailbox providers want to see.

Automated warmup costs anywhere from $25 to $100+ per month per inbox. For a single inbox, that's reasonable. For an agent managing ten inboxes, the cost adds up fast.

Manual warmup is free but time-consuming. You (or your agent) send real emails to real people who actually engage with them. This works well if your agent has a genuine use case that involves back-and-forth conversation, like customer support or scheduling. The engagement happens naturally as part of the work.

The middle ground is building warmup logic directly into your agent's workflow. Start with low daily limits that increase over time. Prioritize sending to recipients who are likely to engage. Track bounce rates and throttle back if they spike. This is more engineering work upfront, but it means your agent handles its own reputation without depending on a third-party warmup service.

Practical warmup strategy for agent-managed inboxes#

If you're deploying an agent that needs to send email, here's what I'd recommend:

Days 1-7: Cap sends at 10 per day. Only send to addresses you know are valid (no cold outreach yet). If your agent receives replies, respond to them. Every reply is a positive signal.

Days 8-14: Increase to 25-30 per day. Start mixing in new recipients but keep the ratio weighted toward contacts who've engaged before.

Days 15-30: Ramp to 50-100 per day, monitoring bounce rates the whole time. If bounces exceed 2%, slow down. If spam complaints appear, stop and investigate.

After 30 days: Your domain should have enough reputation to handle normal sending volumes. But "normal" depends on your use case. A transactional agent sending order confirmations can handle higher volumes than a cold outreach agent, because transactional emails get opened at much higher rates.

Throughout this process, monitor your domain reputation using Google Postmaster Tools (for Gmail delivery) and check your authentication pass rates. If your SPF or DKIM is failing, fix that before worrying about volume.

When warmup isn't the real problem#

Sometimes emails land in spam and warmup gets blamed, but the actual cause is something else entirely. Content that looks like spam (all caps, excessive links, aggressive sales language) will get filtered regardless of your domain reputation. Sending to purchased email lists full of invalid addresses will spike your bounce rate and destroy your reputation overnight. And if your "from" address doesn't match your authentication records, no amount of warmup will save you.

Warmup is one piece of a larger puzzle. It handles the reputation-building part, but it can't compensate for broken infrastructure, bad content, or dirty recipient lists.

For AI agents in particular, the biggest risk isn't skipping warmup. It's the agent not knowing warmup exists as a concept. If you're building agents that send email, build the sending limits and ramp-up logic into the agent's configuration from the start. Don't let it discover deliverability problems after the domain is already burned.