Amazon blocks Meta's Muse: the engineering problem AI agents create

In this week’s episode of Keeping up with the AI Agents, we have: Meta's Muse.

What happened between Amazon and Meta

Meta built an AI assistant that can perform tasks on a user's behalf, including shopping online. Amazon has now blocked Muse from using Amazon.com, saying the agent was accessing the site without authorization and raising security and transparency concerns. Amazon says it had asked Meta to remove Amazon from the experience before taking action.

Meta disputes some of Amazon's security claims. It says Muse cannot see users' passwords or payment information and that credentials are kept separately from the model.

Software was built for a human at the screen

For years, software was designed around a pretty simple assumption: a human is sitting in front of the screen.

Then came the bots, and now those bots have bots of their own.

According to Amazon’s user journey: a human searches Amazon, compares products, selects one and clicks checkout.

Now you have software doing that, and it’s an engineering problem.

An agent needs identity. It needs permissions. It needs to know what it is allowed to access. It needs to understand when a website has said no. It needs to handle credentials without exposing them to the model. It needs to recover from errors without making an even bigger mistake.

And someone has to build all of that.

Who builds the guardrails

If you’ve been keeping up with our content, you know the importance of building the infrastructure that makes autonomous systems trustworthy.

Identity engineers are going to have to think about non-human identities.

Security engineers will have to understand what happens when software can make decisions and take actions.

Platform engineers will have to create controlled environments where agents can operate without having unrestricted access to a company's systems.

Infrastructure leaders will have to figure out how thousands of autonomous processes can run reliably without creating a security nightmare.

And product leaders will have to decide which decisions an agent can make independently and which ones still require a human.

The Amazon-Meta fight is just one example, but the underlying problem is going to show up everywhere.

Banks won't want an unknown agent moving money.

Healthcare companies won't want an agent wandering through patient systems.

Enterprises won't want an AI assistant with unlimited access to internal applications.

And websites won't necessarily want somebody else's agent controlling the customer relationship.

That means companies are going to spend heavily on the layer between AI capability and real-world action.

If you're a senior technology leader thinking about where your career is heading, we can share what companies are and aren’t looking for in your target roles.

Heck, we can even land you those roles.