What happened
Walmart just did what almost no other big retailer is willing to do: it rolled out the welcome mat for AI shopping agents — and handed shoppers a leash to go with it.
On Wednesday, October 21, 2026, at the Masters of Scale Summit in San Francisco, Walmart president and CEO John Furner announced Agent Gateway, a new system that lets approved AI agents connect to Walmart's online store and actually buy things on your behalf. The same setup covers Sam's Club, the company's discount warehouse.
The timing is the story. Walmart's biggest rival, Amazon, recently started blocking AI agents from buying on behalf of shoppers. Travel and booking sites have done the same, slamming the door on software that tries to shop for people. Walmart went the other way.
The plug socket behind it all
Agent Gateway is built on something called an MCP server. MCP stands for Model Context Protocol, and Anthropic — the company behind the Claude chatbot — created it. In plain terms, it's a universal plug socket for AI. Instead of every retailer building a custom pipe for every chatbot, the agent plugs into one standard socket, searches inventory and completes a purchase.
Walmart already had partnerships in this space, including a deal with Meta's Muse, plus integrations with ChatGPT and Gemini. Agent Gateway is the plumbing that makes those partnerships real.
Agent Profiles: a debit card for your chatbot
Starting next month, Walmart rolls out Agent Profiles. Think of it exactly like adding an authorized user to your credit card: your kid gets a card with their name on it, you see their spending separately from yours, and you set a cap.
Here's what that looks like in practice:
• Every AI agent you use gets its own profile tied to your Walmart account.
• Its purchases show up separately from your own.
• You choose how to approve things: sign off on every single transaction, or pre-authorize spending up to a limit you set.
• You can delete an Agent Profile whenever you want.
• You don't need Walmart+ — any Walmart account works.
Walmart also wrote a set of terms for agents themselves, spelling out how they're allowed to behave and what rights they have to transact. And the returns policy still applies, including when an agent doesn't do what it was told.
What it means for you
An AI agent, if you haven't bumped into one yet, is a chatbot with a wallet. You describe a goal — "restock the coffee, the oat milk and the dog food, keep it under $80" — and instead of handing you links, it goes and does the job.
Until now, that was mostly a demo. Walmart is turning it into a checkout button. Here's where it lands in real life.
At home: the grocery run that runs itself
Household shopping is the killer use case because it's boring, repetitive and easy to describe. Your agent reorders the same brands you always buy — and Walmart says it can do that without the agent ever seeing your personal details or purchase history. You approve each order, or set a monthly cap and let it run.
At work: the supply closet that refills itself
Offices run on the same 20 items: printer paper, coffee pods, cleaning wipes, cables. An agent with a monthly limit you set can keep that shelf stocked without anyone filling out a form. Because agent purchases are tagged separately, finance sees "agent spend" as its own line instead of mystery charges.
For a small business: restocking without a spreadsheet
If you sell online or run a shop, low-stock panic is a weekly ritual. An agent can watch your list of consumables and reorder before you run out. The spending cap is the important part — it's the difference between a helpful assistant and a surprise invoice.
For students: dorm supplies on a fixed budget
A student on a tight dorm budget can ask an agent to kit out a room and never risk an overdraft, because the cap is set before the agent ever shops. Per-purchase approval means parents can stay in the loop without hovering over every click.
For creators: gear restocks and consumables
Ring lights burn out. SD cards fill up. Gaffer tape disappears. An agent that knows your kit list and your ceiling price is a genuinely useful studio assistant — and you keep the receipts separate from your personal spending.
For extra income: sourcing without the tab-hopping
Anyone reselling or flipping products lives on price checks and restocks. Agents that can search inventory and buy within a limit cut the boring half of that work. It's not a money printer; it's a faster pair of hands.
If you want to see how agents and other AI shortcuts fit into a small operation before you hand one your card, there are free tools worth bookmarking at MyKreaTool.
How to try it right now
You can't flip a switch on everything today — Agent Profiles land next month — but you can line up the pieces now, and the cheapest route costs nothing.
1. Start free: open a Walmart account. No membership, no fee. A basic account at walmart.com is all that agent support requires, which makes this the cheapest experiment in agentic shopping you'll find anywhere.
2. Pick a free agent. ChatGPT and Gemini both have free tiers, and Walmart already offers integrations with them. Use the free version first — you're testing the workflow, not the model.
3. Write the prompt like a shopping list with rules. Something like: "Find the store-brand version of my usual paper towels, laundry detergent and dish soap. Total must stay under $70. Show me the cart before buying." Specific brands, a hard number, and a confirmation step.
4. Wait for Agent Profiles, then set your leash. When they roll out next month, a profile is created the first time you shop through Walmart's Agent Gateway with a given agent. You authorize it, link it to your Walmart account, then pick per-purchase approval or a spending limit.
5. Test small. Run one $20 order with per-purchase approval switched on. You'll learn more from one real transaction than from ten articles — including this one.
6. Check the separation. Confirm the agent's purchase shows up as its own line in your account. That's the feature that makes this safe to scale.
Upsides and what changes
The obvious win is time. Recurring shopping is a 20-minute chore that an agent can compress into a 10-second approval — and Walmart is betting that shoppers who can hand off the boring part will spend more, not less.
The second win is accountability. Tagged agent purchases, spending caps and per-transaction approval aren't glamorous, but they're exactly what people need before they let software spend their money. The credit-card analogy is a good one: it's a familiar mental model, not a new thing to learn.
The third is competitive. By opening up while Amazon closes down, Walmart positions itself as the default store for anyone who shops through an AI assistant. That's a land grab, and it's happening right now.
Limitations
Be clear-eyed about this one. Agent Profiles don't exist yet — they're "next month," which means everything above is a plan with a date on it, not a feature you can use this afternoon. The MCP server only works with approved agents, so not every chatbot you like will be allowed in. There's a real learning curve too: writing a prompt precise enough to get the right brand, the right size and the right price takes a few tries, and an agent that grabs the 48-pack instead of the 12-pack is annoying even when the return is free. Walmart says its returns process covers cases where an agent doesn't act as intended, which helps — but you're still the one packing the box. And the whole thing assumes you're comfortable handing purchasing power to software in the first place. Plenty of people aren't, and no spending cap changes that feeling overnight.
Conclusion
Walmart just made AI shopping agents a normal thing to try rather than a sketchy experiment — with a separate profile, a spending cap and a yes/no button standing between the bot and your bank account. It's a direct contrast to Amazon's block-and-ban approach, and it puts a usable version of agentic commerce in front of ordinary shoppers first.
Your one action today: open a free Walmart account and write the prompt you'd use for your most repetitive order — brands, sizes, and a hard dollar limit. Don't connect anything yet. Just see how precisely you can describe the job. That prompt is the whole skill.
Comments 0