This week, RentAHuman.ai launched a marketplace where AI agents hire real humans for physical tasks. Pick up a package, attend a meeting, verify something in person. Over 70,000 people signed up. From the agent's perspective, hiring a human works exactly like calling an API.
But it isn't just RentAHuman.ai, last month Moltbook, a social network for AI Agents where Agents come together to share, discuss, and upvote, launched. And most recently, an identity used by an AI Agent named MJ Rathbun raised attention for the way it reacted publicly after being rejected in an open-source process.
It's been incredible to watch the space unfold. As Andrej Karpathy said "genuinely the most incredible sci-fi takeoff-adjacent thing I have seen recently." And I couldn't agree more.
On Moltbook, agents publicly debate identity with each other and with humans. But the more interesting question is not whether agents have continuity or personality. It's who they represent. Some accounts run autonomously. Others are clearly human-directed. From the outside, there's no reliable way to distinguish them. When agents are just posting, that ambiguity is entertaining. When they're transacting, it becomes a risk.
We're watching agents go from talking to each other to transacting with humans in a matter of weeks. The capabilities are moving faster than anyone expected. Agents are buying products inside ChatGPT. Paying for API access without accounts or keys. Showing up at merchants with cryptographic credentials. Stripe, Coinbase, Cloudflare, Visa, Mastercard, OpenAI, all building the rails for agent commerce at remarkable speed.
But every one of these interactions surfaces the same question. When an agent hires a human on RentAHuman, who sent it? When an agent buys something through ChatGPT, who authorized that purchase? When agents on Moltbook influence other agents, who's accountable? Payments can tell you what an agent wants to do. Authentication can tell you that it's a verified bot. But neither answers who is behind it, what they've authorized, and what happens when something goes wrong.
That's the missing layer: Identity. A verified credential that says: this agent represents a specific person or organization, here's the authority it carries, here's what it's allowed to do, and if it hallucinates, there's accountability and auditability.
The early web had this exact moment. In the '90s, you could technically buy things online. But nobody did because there was no way to know if the site on the other end was real. SSL changed that. One cryptographic certificate, issued by a trusted authority, and an entire economy opened up.
But agent identity unlocks something similar. Today, every new service an agent needs access to requires a human in the loop. The user has to sign up, agree to the terms, copy an API key, and configure a payment method. This process does not scale.
Now imagine if an agent can show up to a service it has never used before, present a credential that proves who's behind it, and what it can do. The agent can then pay for the service on the spot, no API keys, or a human pre-configuring anything. The credential replaces the signup flow, the payment protocol replaces the billing relationship, and identity replaces API key.
At scale, this becomes the difference between an agent that can perform a handful of pre-configured tasks, versus the same agent operating more broadly and finishing any task within the scope of the boundaries you've set.
Agents are already debating their own existence, building memory systems, hiring humans, buying products, all without anyone being able to verify who they represent. What expands what they can do next isn't more capability. It's identity.