AI Agents Can Now Take Orders Through Chat. Can They Follow Up By Email?

On July 1, 2026, Square announced that eligible sellers can now take orders directly inside ChatGPT and Claude. A customer can ask an AI assistant where to get dinner, see a nearby restaurant's menu, and place the order — without ever leaving the conversation. It is the clearest sign yet that agentic commerce has crossed a line: AI agents are no longer just answering questions about products. They are closing transactions. But a transaction is not the end of a relationship. The receipt, the order confirmation, the "your food is being prepared" update, the refund when something goes wrong — all of that still runs on email. And most agents that can now take an order have no inbox to send it from.

What Square Actually Shipped

Square's July 2026 launch lets eligible U.S. food-and-beverage sellers with an active Square Online Ordering profile appear inside AI conversations and accept orders there. The integration ships as a ChatGPT app on OpenAI's side and a Claude plugin on Anthropic's, with Amazon Alexa+ voice ordering described as in development. Orders placed in the chat flow straight into the seller's existing Square stack — Point of Sale, Kitchen Display System, and reporting — so the restaurant runs the order exactly as if it came through their own website.

Two details matter for anyone building on top of this. First, eligible sellers were enrolled automatically, with what Square described as no additional work, setup, or fees, and no new marketplace commission on AI-channel orders — a pointed contrast with delivery platforms that take a cut. Second, the plumbing is Square's Universal Commerce Protocol, co-developed with Google, not OpenAI's checkout protocol. Different platforms are converging on the same behavior — buy inside the assistant — through different standards.

Morgan Kuntze, who leads global partnerships at Square's parent company Block, framed it plainly: "Modern commerce is moving at a sprint, and we're building Square to help sellers appear everywhere customers are going." Square cited its own supporting data alongside the launch: that 42% of consumers now use AI tools for shopping, and that agentic shoppers could drive roughly $385 billion in U.S. ecommerce by 2030. Square reaches about 4.5 million sellers across channels; the food-and-beverage cohort eligible for AI ordering is a subset it did not size.

This Didn't Come Out of Nowhere

Square's announcement is the retail-storefront edition of a shift that started nine months earlier. Here is the timeline that got us here:

The load-bearing piece under most of this is the Agentic Commerce Protocol (ACP), the open standard OpenAI and Stripe published on September 29, 2025. ACP uses a Shared Payment Token — a credential scoped to one merchant and one cart total — that the agent hands to the merchant, who charges it through their own payment provider. Crucially, the merchant stays the merchant of record: they accept or decline the charge, own fraud and tax, and handle fulfillment. Stripe's Kevin Miller put the shift in one line: "Stripe has spent the last 15 years optimizing commerce for human buyers. Now, we are starting to do the same for agents."

It is worth noting that the model is still being tuned. In March 2026, OpenAI reworked its shopping experience to lean more on product discovery and route some buyers back to the merchant's own checkout, saying the first in-chat version "did not offer the level of flexibility that we aspire to provide." The direction is set; the exact choreography of where the final click happens is still moving.

The Money Says This Is Real

Agentic commerce is not a demo anymore — the measured numbers from the 2025 holiday season prove buyers are already arriving through AI. Adobe Analytics reported that traffic to U.S. retail sites from generative-AI sources rose 693% year over year over the holidays, and those visitors converted at a meaningfully higher rate than the average shopper. Salesforce estimated that AI and agents influenced roughly 20% of all retail sales — about $262 billion — over the same global holiday window.

The forecasts point the same way. McKinsey projects agentic commerce could orchestrate up to $1 trillion in U.S. retail revenue by 2030, and $3–5 trillion globally. Morgan Stanley — the source behind Square's $385 billion figure — puts agentic shoppers at 10–20% of U.S. ecommerce by the end of the decade. You can quibble with any single projection, but the direction is not in dispute. A growing share of transactions will begin, and increasingly end, inside an AI conversation.

The Gap Nobody Is Budgeting For

Here is the problem: everything above is about the moment money changes hands. Almost none of it is about what happens next. And what happens next is where customer relationships are actually made or lost.

Think about the last thing you bought online. The purchase took ten seconds. Then came the order confirmation, the receipt, the "your order has shipped" email with a tracking link, maybe a delivery confirmation, possibly a support thread when the size was wrong, and eventually a "how did we do?" follow-up. That is the relationship. It plays out over days or weeks, and every message in it lands in an inbox — not a chat window.

This is not a preference we are asserting; it is what consumers say they want. Sinch's research on transactional messaging found email is the single most-preferred channel for every category that matters after a sale:

And people don't just tolerate these emails — they open them. Post-purchase flows like order and shipping confirmations open at roughly 59–60% in Klaviyo's benchmarks, about double the average automated email and far above any marketing campaign. Transactional email is the highest-engagement mail a business sends. It is the exact opposite of a channel you want to skip.

So the question for anyone building a commerce agent is uncomfortable and specific: your agent can take the order — but what address does the confirmation come from? If the answer is "the chat platform handles it" or "the merchant's existing system," then your agent is a front end for someone else's relationship. The moment you want the agent itself to own the post-purchase experience — to send the receipt, answer the "where's my order?" reply, and follow up next week — it needs an inbox.

Why Chat Can't Do the Follow-Up

Chat is the wrong channel for the post-transaction lifecycle for a structural reason, not a stylistic one: chat sessions are ephemeral and unaddressable; email is persistent and addressable.

When a ChatGPT conversation ends, there is no durable thread the merchant can reach back into three days later to say "your order shipped." There is no address the buyer can reply to when they have a question at 11pm. The assistant may have moved on to a hundred other tasks. Email solves this by design: every message has a stable address and a threadable identity (the Message-ID, In-Reply-To, and References headers), so a conversation can pause for a week and resume exactly where it left off. That is also why email works so well as durable memory for long-running agents — the same properties that make it good memory make it good customer communication.

There is a second reason. Email is the identity layer of the internet. Somewhere between 65% and 92% of web applications use the email address as the primary account identifier, which is why order confirmations, password resets, and receipts are all keyed to it. An agent that transacts on a customer's behalf is participating in a system whose system-of-record is email. Giving the agent its own address makes it a first-class participant in that system instead of a guest in someone else's session. We made the fuller version of this argument in why agent identity has to include email.

What a Complete Commerce Agent Looks Like

The complete version is not complicated. It is the checkout you already built, plus a mailbox the agent owns. Here is the shape of it with Dead Simple Email:

import requests

API_KEY = "dse_your_api_key"
BASE = "https://api.deadsimple.email/v1"
headers = {"Authorization": f"Bearer {API_KEY}"}

# The agent already closed the sale in chat. Now it owns what happens next.
# 1. Give the ordering agent its own inbox.
inbox = requests.post(f"{BASE}/inboxes", headers=headers, json={
    "name": "orders",
}).json()   # -> orders@yourshop.deadsimple.email

# 2. Send the order confirmation the moment checkout completes.
requests.post(f"{BASE}/send", headers=headers, json={
    "from": inbox["email"],
    "to": customer_email,
    "subject": f"Order {order_id} confirmed — thanks!",
    "text": "Your order is in. We'll email a tracking link when it ships.",
})

# 3. When the customer replies "where's my order?", the agent gets it as JSON.
#    Webhook payload lands at your endpoint the instant the reply arrives.

The customer's reply arrives at your webhook as structured JSON, so the same agent that took the order can read "where's my order?" and answer it — in the thread, from the same address, with no human in the loop and no OAuth. That is the difference between an agent that can sell and an agent that can serve.

Where This Leaves Builders

Agentic commerce spent 2025 and 2026 solving the hard, glamorous problem: how does an AI safely move money on a person's behalf? Square, OpenAI, Stripe, Visa, Mastercard, Amex, and PayPal have largely answered it. The unglamorous problem — who sends the receipt, who answers the reply, who follows up — is still sitting there, and it is the part customers actually experience most.

If you are building a commerce agent, treat the inbox as part of the checkout, not an afterthought. The order confirmation is not a nice-to-have; it is the first message of the relationship, it is the mail people are most likely to open, and it has to come from somewhere. Give your agent an address of its own and it can own the entire arc — discover, transact, confirm, support, and follow up — instead of handing the best part to someone else.

Frequently Asked Questions

What is agentic commerce?

Agentic commerce is a shopping model where an AI agent discovers products, compares options, and completes a purchase on the buyer's behalf inside an assistant like ChatGPT, Claude, or Gemini — without the buyer visiting a store website. The 2026 rails include OpenAI and Stripe's Agentic Commerce Protocol, Square's ChatGPT and Claude ordering integrations, and agent-payment programs from Visa, Mastercard, American Express, and PayPal.

Can AI agents actually complete purchases now?

Yes. OpenAI's Instant Checkout launched with Etsy and Shopify in September 2025, Walmart joined in October 2025, and Square extended in-chat ordering to food-and-beverage sellers in July 2026. Payment runs on scoped, single-transaction tokens, and the merchant remains the merchant of record for fraud, tax, and fulfillment.

Why does a transacting AI agent still need email?

Because the checkout is one moment and the relationship is a lifecycle. Order confirmations, receipts, shipping and delivery notifications, support replies, and follow-ups all run on email — the channel consumers prefer for each of those message types. A chat session is ephemeral and has no address to reply to; email is persistent and addressable, so the agent can own everything that happens after the sale.

Isn't the chat platform or the merchant already sending the confirmation?

Sometimes — and if so, your agent is a front end for someone else's relationship. The moment you want the agent itself to send the receipt, answer the buyer's reply, and follow up later under its own identity, it needs its own inbox. Otherwise the most-opened, highest-trust messages of the entire journey belong to another system.

How do you give an AI agent its own inbox?

With an agent-native email API like Dead Simple Email, you create an inbox with one API call, send confirmations and updates through it, and receive replies as structured JSON via webhooks — no OAuth, no human in the loop. Pricing starts free for 5 inboxes and is $29/mo for 100 inboxes with 100,000 emails per month. See the full cost comparison for how that stacks up against other options.