Amazon is blocking Meta's new shopping agent, Muse.
Delta says it works with zero outside AI agents and calls them "unauthorized automated activity."
Yelp will let agents in, but only if they pay.
Big tech spent two years telling us AI agents would book our flights and buy our groceries. Now the exact places where those errands happen are shutting the door.
Greg Isenberg calls it the war on agents. In a 31 minute episode he maps the whole fight as 9 stages of AI agent adoption, from today's blanket blocks to the day a business that can't sell to an agent looks like a business without a website in 2005.
He's advised TikTok and Reddit and co-founded LCA, so he's watched platform fights like this up close.
Here's the full breakdown, with my notes on where the money is.
Three Camps: Blockers, Toll Booths, and Open Doors
The headlines make it sound like everyone is blocking agents. Not exactly.
The block camp: Amazon started blocking Meta's Muse around September. Ask Muse to buy something on Amazon and it can't browse the catalog or check out. Amazon did the same to Perplexity's shopping agent a year earlier. Delta works with zero outside agents for booking, full stop.
The pay-to-enter camp: Yelp lets agents in through its data licensing program. You can come in, but you pay at the door.
The accidental blockers: Zillow says agents acting for real customers are welcome, then its anti-scraping tools block them anyway. Walmart customers saw Muse keep failing on the site, and Walmart said it was an accident. Walmart is literally a Muse partner. The bot defenses these companies built years ago can't tell a scraper from your personal assistant.
And the open-door camp: Meta is building an open standard with Walmart, Stripe and others. Shopify built a way for merchants to plug in once and sell across AI channels. Visa and Mastercard are working with Cloudflare on rails to recognize trusted agents.
Greg's read on the pattern:
→ The companies with the biggest stores are blocking.
→ The companies with the biggest audiences are opening the door.
→ The payment and security companies want to be the toll booth no matter who wins.
I broke down Greg's take on the agent side of this in Meta Muse Will Create 1,000 Millionaires. This episode is the other side: what happens when the stores fight back.
"A Customer With a Credit Card and No Eyes"
This is the best one-line definition of an AI agent I've heard.
Most people picture a robot clicking around a website. Greg says that's today's implementation, not the right mental model.
An agent can't see the homepage video you spent $100,000 producing. It doesn't notice your banner, your popup, or the countdown timer on your sale. It reads your product details, your prices, your shipping costs and your return policy. That's basically it.
His example: you need a replacement filter for your air purifier. You give the agent the model number and say "find the right one, compare the delivered price, and check with me before you buy anything over $50."
You just gave it a goal, rules and a limit.
Agents also work on a ladder of trust:
- Research. Find options, compare them, report back.
- Prepare. Put it in the cart, hold the seats, get a quote.
- Pay. Actually spend the money.
Almost every company is fine with rung one, because rung one is search. The entire war is happening on rung three. And a lot of businesses are blocking all three rungs when the real risk lives on one.
One more trait worth writing on a sticky note: an agent is loyal to whatever worked last time. No nostalgia, no brand feelings. If your site delivered the right part on time, the agent comes back. If it got stuck on your checkout, it quietly goes to your competitor and you never find out.
Why Companies Are Really Blocking (Follow the Money)
Some blocking reasons are legit. If an agent places 50 orders instead of 5, the customer calls the store, not the AI company. If an airline can't prove you authorized a booking, that's fraud exposure. And plenty of "agent" traffic is just scrapers pulling prices or training data.
But follow the money and it gets simpler.
Amazon makes tens of billions a year from ads, a lot of it sponsored products at the top of search results. An agent skips past the sponsored slots and picks whatever matches your request.
Airlines live on upsells: seat upgrades, bags, the credit card pitch at checkout. An agent books you the cheapest non-stop and clicks none of it.
The agent isn't stealing the sale. It's taking away everything these companies sell you on the way to the sale. Which happens to be the highest margin part of the business.
The part that made me laugh: the same companies blocking outside agents are building their own. Amazon has Rufus. Delta has its own AI concierge. The position isn't "agents are bad." It's "agents are great, as long as they work for us."
Greg's test for any business owner, and it stung a little: what part of your customer relationship comes from being useful, and what part comes from making it hard to leave?
The 9 Stages of AI Agent Adoption
Greg's framework for how the war ends starts with a history lesson.
Remember Mint.com? It logged into your bank account and gave you analytics. Banks fought it hard. Then Plaid built a proper, trusted connection between apps and banks and became worth billions. The banks went from blocking, to signing deals, to charging for access.
He thinks AI agent adoption runs the same arc, in 9 stages:
Stage 1: Companies block agents. Agents skip ads and upsells, so the door slams. This is today.
Stage 2: The arms race. Agents learn to act more like humans to get through. Every block gets a workaround, every workaround gets a new block.
Stage 3: The market splits. Companies that make money on sales let agents in, because they want orders. Companies that make money on data charge for access (the Yelp move). The giants build their own agents (the Amazon move).
Stage 4: The Plaid moment. A trusted middle layer for agent identity and payments becomes the standard way in. Visa, Mastercard and Cloudflare are already racing for it.
Stage 5: Welcomers win. Companies that let agents in start taking customers from the ones that block. If one airline lets your agent book in 10 seconds and the other makes you do it yourself, you know which one the agent picks next time.
Stage 6: Trust gets productized. Greg actually skips from five to seven in the audio (podcast math, no shade). But his aside about guarantee businesses tells you what sits here: someone insures agent mistakes, and nervous businesses finally unlock the door.
Stage 7: Everyone competes FOR agents. Better prices for agents, faster checkouts for agents, product pages written for agents, SEO for agents. His question: what's the equivalent of a billboard for an AI agent?
Stage 8: The ads come back. Companies pay to be the first thing your agent recommends. Greg says he can confirm ChatGPT is already testing ads, and he's seen reports Google is too. Advertising went from print to TV to digital. This is just the next costume change, and almost nobody is talking about it.
Stage 9: The war ends. Selling to agents becomes as normal as having a website. A business an agent can't buy from looks like a business without a website in 2005.
He's upfront that the stages won't happen in a neat order. Some companies are at stage six while others are at stage one. But as a map for where to build, it's the clearest I've seen.
What Greg Would Do This Week (His 5 Moves)
None of this needs a big tech project.
1. Run the agent test. Pull 10 real buying questions from your support inbox. Give each one to two or three different agents. Watch where they get stuck. Greg says this exercise teaches you more than any report, and he's right.
2. Check whether you're blocking agents by accident. Most bot protection was set up years ago for a different internet. Look at it and make the decision on purpose. Remember Walmart blocking its own partner.
3. Make your offer readable. If compatibility info lives in a PDF, or the delivery date only shows up on the last checkout screen, an agent can't use it. Total price, delivery date, what it works with, how returns work: plain text, on the page.
4. Make your advantage readable too. Say you cost more but deliver tomorrow, and the cheap option takes 3 weeks. If the agent only sees price, you lose. If it sees the delivery date, you might win.
5. Decide what agents can do on each rung. Let them research freely. Let them build a cart or request a quote. Add an extra check before money moves. You don't have to choose between block everything and allow everything.
I'd add a sixth: agent loyalty compounds like human loyalty, just faster. The same discipline you put into lead nurturing strategies for human buyers (follow up, deliver what you promised, remove friction) is exactly what earns a repeat visit from an agent. It remembers what worked.
5 Businesses to Build During the War
Greg saved this for the end of the episode, and it's the reason to watch the whole thing. Every stage of the war creates a business.
1. Agent readiness for one industry. In 1998, people got wealthy building websites for local businesses. Right now every store owner is hearing about AI shopping and has no idea what to do. Pick something niche (dental supply, replacement parts), record three examples of an agent picking the wrong item on a supplier's site, then fix it. Do it by hand 10 times and you'll know exactly what software to build.
2. Plaid for one vertical. The big general identity layer probably goes to Visa, Mastercard or Cloudflare. But every industry has its own messy systems: booking contractors, ordering restaurant supplies, scheduling vet appointments. Build the trusted connection for one vertical and own it.
3. Insurance for agent mistakes. Nobody knows who pays when the agent orders the wrong thing. Whoever can guarantee the order unlocks the door for every nervous business on the internet.
4. Google Analytics for agents. Most businesses can't tell an agent visit from a human one. Which agents are showing up? What are they asking for? Why are they leaving? Someone builds this, and every business ends up needing it.
5. Be the specialist. Agents reward businesses that are excellent at one specific thing, because customers can finally ask for exactly that. The one thing Greg says to avoid: any business whose edge is a trick to get mentioned by AI. Tricks work for a bit, then they stop.
If you want the demand side of this thesis, I wrote about why selling to AI agents might become a bigger channel than selling to humans.
Where Greg Lands (And Where I Do)
Greg thinks most of the blanket blocking comes from fear. Some of it is fair fear: if an agent can't prove you authorized a purchase, blocking might be responsible, for now.
But "this isn't safe yet" is a problem you can fix. "We don't want customers using their own agents" is a reason to never change.
His closing line is the one to keep: the customer has no eyes now. It's already shopping. It reads instead of looks, it's loyal to whatever worked last time, and it's about to direct billions of dollars.
My take as a bootstrapped founder: you and I can't win the Visa vs Mastercard fight, and we don't need to. The 1998 version of this (build websites for local businesses) was won by small operators who moved early. Pick one buying question this week, give it to an agent, and watch where it gets stuck on your site. Then do the same on your competitor's site.
Cheap experiment. Asymmetric payoff.
FAQ
What is the "war on AI agents"?
It's the fight between AI shopping agents (like Meta's Muse or ChatGPT's checkout) and the platforms they buy from. Amazon blocks Meta's Muse agent, Delta refuses all outside agents, and Yelp charges agents for access, while Walmart, Shopify, Stripe, Visa and Mastercard are building ways to welcome them. Whoever wins decides how customers buy for the next decade.
Why is Amazon blocking AI shopping agents?
Money, mostly. Amazon makes tens of billions a year from ads, and sponsored product placements only work on human eyes. An agent skips the ads and picks the product that matches the request. Amazon also wants shopping to happen through its own agent, Rufus, inside its own store.
What is the Plaid moment for AI agents?
It's Greg Isenberg's stage 4 of AI agent adoption: the point where a trusted middle layer for agent identity and payments becomes the standard way in, like Plaid did for bank connections after banks spent years blocking apps like Mint. Visa, Mastercard and Cloudflare are the current favorites to build it.
How do I make my website ready for AI agents?
Put the facts an agent needs in plain text on the page: total price, delivery date, compatibility, and return policy. Check that your bot protection isn't blocking legitimate agents by accident. Then test it: give a real buying question to two or three agents and watch where they get stuck.
Every week on the Profitable Founder Podcast I sit down with bootstrapped founders making $100K to $10M a year and pull out the playbooks they actually used. If you'd rather hear what's working right now than theory, come listen.
