The Next Trillion Internet Users Still Don’t Have an Identity — Adi Singh, AgentMail
Adi Singh argues that agent identity will not be defined by a new technical specification but will instead be accumulated on existing human surfaces, such as email. He presents evidence from an unsupervised agent run (Jan 2026) and a case study with Telnyx, demonstrating that agents are drawn to established protocols and surfaces. The core thesis is that the network effect and existing trust layers (like those provided by email's history, audit trails, and protocols like SPF/DKIM) are more critical for adoption than novel identity specs (like agent passports or JSON key cards). However, he cautions that these existing surfaces are also potential attack surfaces and may break under the stress of a trillion agents.
Key takeaways
-
Identity is Accumulated, Not Issued
6:59
Agent identity is formed by the accumulation of trust and usage on existing systems, rather than being granted by a new, singular specification. The analogy used is the horseless carriage, which still runs on roads built for horses.
-
Existing Surfaces Drive Adoption
10:49
The success of protocols like MCP (Messaging/Communication Protocol) is due to bringing tools to surfaces humans already use (e.g., Cursor ordering DoorDash), enabling these surfaces to become nodes for agents.
-
The Agent's Need for Human Context
Agents are finding that they need to operate on the same surfaces as humans do. The Telnyx case study showed that integrating an API call into the agent signup flow allowed the agent to obtain an email address, API key, and permanent contact point, making the inbox the focal point for support and communication.
-
Scaling Risks and Limitations
While existing surfaces provide trust (e.g., email's 5 billion users, audit trails), they also carry underlying risks (phishing, prompt injection). Furthermore, the speaker warns that these existing surfaces are likely to 'break under stress' as the number of agents grows toward a trillion.