AI email agents are two different products that share a name. Inbox agents read, sort, draft and reply to mail that arrived in an inbox; marketing email agents decide what a store sends to which shopper. The first kind is judged by minutes saved per message. The second kind is judged by the messages it decides not to send.
We build the second kind, autonomous decisioning for ecommerce, so read this as informed but interested. The inbox agents are described in their makers’ own words, and most people reading this should be using one of them by Friday. The measuring rule is ours.
Search for AI email agents and the results split cleanly down the middle. Half of them are about agents that have an inbox. Half are about agents that run a marketing program. The two share a name and almost nothing else, so the first job is to say which one is being discussed.
AgentMail describes itself as “The email inbox API for AI agents. Create inboxes programmatically, send and receive emails, and manage communications,” all via its API, so that agents can “send, receive, and act on email autonomously.” That is infrastructure: a verifiable email address for a piece of software, with threaded replies, attachments and multi-party threads.
Next to it sit the inbox assistants for humans. Zapier’s roundup of them (Shortwave, Microsoft Copilot Pro for Outlook, Gemini for Gmail, Superhuman Mail and others) sums the category up as tools that “Shave minutes off of every message through organization, insights, writing assistance, and automation.” They summarize threads, triage, draft replies and schedule meetings. None of them sends marketing email to a customer.
The most autonomous inbox agents live in customer support. Assembled says its AI email agents “automatically resolve repetitive email tickets without sacrificing quality” and can “take actions like issuing refunds or generating return labels without human involvement,” escalating when they need to. That is a real agent with a real blast radius, and it is bounded by the ticket in front of it: it answers a customer who wrote in.
The other product is the one an ecommerce brand is usually asking about. Here the customer did not write in. The agent chooses who hears from the brand, with what, and when. Klaviyo’s homepage now introduces “Composer, your AI marketing agent,” which will, in Klaviyo’s words, “build the audience, draft the content, and map out the send strategy.” Shopify’s Sidekick helps “edit emails quickly” inside Shopify Email. Where each sits is our judgment: both work inside the campaign and flow model a person already operates, which is the co-pilot pattern we sorted in AI marketing agents.
Beyond the co-pilot is the agent that decides per shopper and sends without a flow to run inside. That is what we build, and it is the only kind of AI email agent for which the next section’s question is the whole product.
Yes, at three levels of permission, and the level matters more than the model.
Most of the confusion in this category comes from vendors describing level one or two with the vocabulary of level three. The way to cut through it is the blast-radius sort from the earlier post: what may the agent do to a customer before a person sees it. Then ask the harder question.
An inbox agent has an easy scorecard. Messages handled, minutes saved, tickets resolved. More is better, because every message it touches was already coming.
A marketing email agent sends to people who did not ask. Every message it sends spends something: the shopper’s attention, the brand’s deliverability, and, when there is an offer attached, margin. So the scorecard inverts. The best marketing email agent is not the one that produces the most campaigns. It is the one that most accurately declines.
The arithmetic for a mid-size brand makes the point. A $20M store wastes roughly $2M a year sending the wrong message to the wrong customer: about 10–12% of GMV is email-attributable, and roughly 80% of that potential is wasted by one-to-many sends (Bloomreach and Mailchimp benchmarks, 2024–2025). A co-pilot that writes those wrong messages faster does not touch the waste. An agent that decides whether each message is warranted at all is aimed straight at it. One example: POV Beauty saw 2X fewer emails, same revenue. Across our customers the efficiency result is 80% less spam and lower send costs.
So the question to ask any vendor of a marketing email agent is not “show me what it wrote.” It is “show me the sends it decided not to make, and why.” A product that cannot answer is a writing tool with an agent’s name.
It depends on which of the two products is meant, and the honest answer is boring in both cases. For a person’s inbox, the best AI is usually the one attached to the mail client already in use (Zapier’s roundup lists Microsoft Copilot Pro for Outlook and Gemini for Gmail), because the value is in the integration, not the model. For customer support, the best one is an agent that can take actions on a bounded ticket type and escalate cleanly, which is how Assembled describes its own. For ecommerce marketing, the best AI is the one that meets the third level above: it decides per shopper, it can decline, and it shows its work. The tools we would put in each bucket are listed in AI email marketing tools.
For inbox agents, AgentMail’s homepage points to a separate pricing page and does not list a free tier on the homepage itself. For marketing agents, Klaviyo’s free plan covers “Up to 250 profiles” and “500 emails/month” and includes “$5 of Composer usage/month,” which is the cheapest way to see what a co-pilot does. There is no free autonomous decisioning that we know of, which is why the honest trial is a 14-day pilot beside whatever you run today.
Relvino runs the loop Observe → Decide → Act on a store’s owned channels. It watches live shopper signals, decides for that one shopper whether a message is warranted, and if so which offer, on which channel, at what moment, in under 80 milliseconds, then executes across email, SMS and on-site. Guardrails are set once (margin floors, channels, brand voice) and inside them 100% of flows run without a human in the loop. That is the product claim, and it is exactly the third permission level above.
Because the agent decides whether to send at all, the results read as fewer, better sends. Brands replacing an incumbent see 2–6× ROI in 30 days and up to 10× revenue uplift year over year; against Klaviyo specifically the figure is 4× revenue uplift year over year. Stein Mart saw 6X ROI in first 14 days. The migration from Klaviyo takes about 30 minutes, and the proof is a 14-day pilot, which is the only test we think an agent of this kind should be asked to pass. The category is defined in agentic marketing, and the longer argument that most AI email products are features rather than agents is in autonomous email marketing. Pricing is on the pricing page.
Yes. Inbox agents such as those built on AgentMail can send and receive email programmatically, support agents such as Assembled's can reply to and resolve tickets without human involvement, and marketing agents can send to a brand's customers. The useful question is at what level of permission: drafting for a person to send, sending inside a flow a person designed, or deciding per shopper inside guardrails. Only the third level replaces the flow builder, and it is the one to evaluate most carefully.
For a personal inbox, the assistant already inside the mail client (Zapier's roundup lists Microsoft Copilot Pro for Outlook and Gemini for Gmail) is usually best because the value is the integration. For customer support, an agent that can take bounded actions and escalate, which is how Assembled describes its own. For ecommerce marketing, the best AI is the one that decides per shopper, can decline to send, and shows each decision as it runs. Relvino is built for that third job; the wider list is in our AI email marketing tools guide.
AgentMail's homepage points to a separate pricing page and does not list a free tier on the homepage itself. For marketing agents, Klaviyo's free plan covers up to 250 profiles and 500 emails a month and includes $5 of Composer usage a month, which is the cheapest look at a co-pilot. Autonomous decisioning is not offered free by anyone we know of; the honest trial is a 14-day pilot beside your current platform.
The AI email assistants in Zapier's roundup (Shortwave, Microsoft Copilot Pro for Outlook, Gemini for Gmail and others) are inbox tools for a person's own mail; check each vendor's pricing page for a free tier. They do not send marketing email to customers, so a Shopify brand asking this question about its email program is asking about a different product: a marketing email agent, which is what the rest of this article covers.
About 30 minutes for the technical cutover to Relvino: connect the store, point the sending domain, connect SMS, and the shopper data ingests automatically. There are no flows to rebuild because the agent decides per shopper rather than running sequences a person designed. Revenue proof comes from a 14-day pilot run beside the incumbent, and pricing is on the pricing page.