RFC 8628 is interesting exactly because it separates the authorization surface from wherever the agent runs. We have been looking at similar patterns. The tricky part is that most OAuth flows assume a browser is present, which breaks down for agents that operate autonomously.
What I find even harder though is the cross-organizational case. Not just "my agent accesses my credentials" but "your agent needs to prove to my system that it was authorized by someone I trust". At that point you need identity and authorization as separate layers, and most current solutions kind of mash them together.
Pods in AgentLair give you namespace isolation (each pod gets its own vault, email, keys), which helps for the multi-tenant case. But the trust problem across organizations is still largely unsolved infrastructure-wise.
But the interesting thing is, my actual token usage running agents is way less than people here seem to assume. Most of the time the agent is waiting for tools, reading files, thinking. The bursts are intense but short. I probably use less tokens per hour than someone doing a long manual coding session with lots of back and forth.
The real issue for me isnt cost, its that they can just change the rules whenever. I had to drop everything today to verify my setup still works. Thats the tax of building on someone elses platform I guess.
Supply chain attacks like the LiteLLM compromise exfiltrate every env var, SSH key, and API key from compromised machines. AgentLair Vault prevents this architecturally — credentials are never in the environment.I built AgentLair to give AI agents a real identity — not just an email address.AgentMail raised $6M to solve the email problem. They've done great work, and email IS the starting...