Twenty-six years ago, I landed in a country I'd never visited but somehow already knew. I thought the split existence I'd live, here and there simultaneously, was just the immigrant condition. Turns out, I was early to something much bigger!
At the end of the 1960s, digital identity had a problem: to verify identity, the system must know the secret; but if the system knows the secret, the secret can be stolen.
If AI generates the content, trains on that content, then deploys agents to navigate that content on your behalf—at what point does your identity become whatever the model consensus decides you are?
The "AI Agents As Employees" paper:
- Promises explainability, admits black boxes
- Calls them teammates, admits they can't reciprocate
- Proposes social contracts with non-agents
- 57 pages, 100 words on identity management
Words matter when liability follows.
Every time you log into a website, unlock your phone, or prove you're not a robot, you're participating in a system that's become dizzyingly complex. Behind those simple moments—typing a password, getting a text code, scanning your face—sits an entire industry of standards and protocols: NHI (non-human identities), IAM (Identity and Access Management), SSO (Single Sign-On), MFA (Multi-Factor Authentication), RBAC (Role-Based Access Control), OAuth (Open Authorization), OIDC (OpenID Connect), SAML (Security Assertion Markup Language), and dozens more acronyms that all relate to proving you are who you say you are and determining what you're allowed to access. How did things become this complicated?
ChatGPT can now buy things for you. Google just released a protocol for AI agents to handle payments. And nobody's asking the obvious question: who's advertising to your AI?
OpenAI's Instant Checkout lets you shop without leaving the chat. Google's Agent Payments Protocol (AP2) gives AI agents the infrastructure to transact on your behalf. Both companies insist results are "organic" and "unsponsored."
Right. Because when you build a payment rail that processes transactions through conversational AI, advertisers definitely won't figure out how to optimize their product descriptions for agent preferences instead of human eyeballs.
We spent decades learning how to game Google's algorithm. How to buy Facebook ads that target your exact neuroses. How to SEO our way into your search results.
Now we're handing purchase decisions to agents that "consider factors like availability, price, quality" and trust them to rank options "purely on relevance."
Relevance to what? The prompt you typed? The preferences they've inferred? The merchants who've optimized their listings for whatever embedding space these models operate in?
The shift isn't humans buying things versus AI buying things. It's discoverable persuasion versus invisible influence. At least when you see an ad, you know someone paid to change your mind. When your agent "recommends" something, you'll assume it's acting in your interest.
This week: Spotify deploys AI to fight AI voice theft while removing 75 million spam tracks. Hollywood recoils as synthetic actress Tilly Norwood shops for agents. Britain mandates digital IDs for all workers by 2029, complete with biometric verification on your phone. OpenAI turns ChatGPT into a checkout counter. California passes its first AI safety law with million-dollar fines. Meanwhile, the iRobot founder says the humanoid robot hype is just that, and radiologists are getting raises despite AI that reads X-rays better than they do. When we can verify everything about a person but recognize nothing—what exactly are we authenticating?
