A developer in the US got tired of the adverts on his parents' Xiaomi television.
He gave Claude Code the model number, let it find the set on the home wifi and watched it disable 25 of the manufacturer's packages.
He posted the result on X, where 8.6 million people saw it.
Within 10 days the project that grew out of that post, CleanYourTV, had more than 41,000 stars on GitHub - a code repository - 600 contributors and tested instructions for around 300 television models from 83 brands.
If you want to do it yourself, the whole job takes about 15 minutes.
The manufacturer's next firmware update puts the adverts back, so the project also ships a tool that reapplies the clean-up afterwards.
I start with the television because it is the small version of a much bigger story.
A customer with an agent is a different customer from one with a remote control. Companies haven't woken up to the power this gives them.
All you have to do is imagine business with no friction.
Muse
Meta launched Muse on 8 September. It is a personal agent that sits inside Facebook, Instagram, WhatsApp and Messenger, runs on its own cloud computer, fills in forms, books travel and pays through a Stripe wallet.
It is free.
On Meta's earnings call on 21 October, Mark Zuckerberg said more than 10 million people in the US and Canada had set one up in six weeks, about a quarter of them use it every day and shopping is the most common thing they do with it, at an average basket of more than $200.
Demand in the first week was high enough that Meta had to ask Stripe to raise its credit limits.
HUMAN Security, which measures bot and agent traffic for websites, found Muse already:
- accounts for about 72 per cent of all AI agent requests it sees
- that 84 per cent of those requests land on product and search pages
- and that Muse is 25 per cent more likely than other agents to reach a checkout page.
It reached in two days the request volume that took ChatGPT 11 months.
In other words, roughly three per cent of North America acquired a shopping agent in six weeks.
Blockers
But there are always blockers in life.
Amazon blocked Muse on 20 September, 12 days after launch, after Meta declined a request to take Amazon out of the agent.
Anyone who tries now sees a message saying an unauthorised agent breaches Amazon's conditions of use. Amazon's robots file already names ChatGPT, Claude, Copilot, Perplexity, Grok and Google's crawlers, and the company is in court with Perplexity over its Comet browser.
Charles Schwab and JPMorgan Chase stop agents touching customer accounts.
Schwab's chief executive says the firm blocks Muse tens of times a week, because nobody can say who is liable when something goes wrong.
eBay rewrote its user agreement to ban unauthorised agents from shopping on a customer's behalf.
Who opened their platforms to Muse?
But some brands have gone all in.
Walmart, Best Buy, Gap and Sephora signed up as Muse partners before launch, and Walmart is building a gateway with agent profiles so approved agents can search its inventory and buy.
QVC and HSN let agents complete purchases and feed their catalogues to OpenAI and Google.
Shopify opened its whole merchant network to agents, on the condition that an agent buying for a shopper includes a human approval step and uses Shopify's checkout.
The threat
Amazon's advertising business is worth about $68bn a year and almost all of it depends on a person seeing a sponsored row.
But an AI agent compares things and commoditises.
It reads prices, stock, delivery terms and returns policies as published and picks.
Sponsored placements mean nothing to agents.
Meta has already said it expects businesses to pay a fee on transactions Muse completes for bringing you a customer you may already have had.
The alternative is to turn the agent away and ask that customer to shop by hand. Either way the customer relationship has a new middleman, and the middleman is whoever owns the agent.
Europe
Muse is unavailable in the UK and the EU. So are OpenAI's and Google's agents. Meta has given no date, and the reason is the rule set it has to meet before an agent can act on a European's email, payments and health data: GDPR and the AI Act's requirements for transparency and human control.
That gives every European business a window of a few months, and the window is being spent as though nothing were happening.
The boards I talk to have a policy for scrapers but often nothing for an agent acting on a customer's explicit instruction with the customer's own credentials.
Your challenges
Three decisions follow:
1 - Decide which kind of store you are, because the default is to be closed by accident
2 - Decide what you will charge or pay, because the agent platforms will set the terms if you leave the seat empty.
3 - And decide what your customer relationship consists of once the customer has stopped visiting, because the agent won't look after that.
I cannot believe how fast this is moving. Between the security vulnerabilities, hacker innovations and new customer controls from agents, my view is that most companies are very far behind in the conversation.
DM me to chat about how to get your business ready.
Dan
