Agent identity will accumulate through existing surfaces such as email instead of arriving fully formed from a new specification.
2
Existing surfaces give agents recognition, history, permissions, and reachability, but they also create attack risks such as phishing, prompt injection, and spam.
3
An AgentMail inbox let Telnyx's autonomous signup flow verify an agent, issue an API key, and keep a permanent contact point for billing, support, and recovery.
Summary
Adi Singh argues that agents will gain identity by using the same surfaces people already use. Email gives an agent an address, a history, a way to receive verification codes, and a permanent contact point. Those properties can build recognition and trust over time, although they also bring phishing, prompt injection, spam, and a larger blast radius when an agent has access to a person's accounts. Singh contrasts this gradual path with identity specifications based on agent cards, passports, and tightly scoped permissions. He says these designs may be technically sound but can restrict adoption. In a Telnyx signup flow, an AgentMail inbox let an autonomous agent register, verify its email, receive an API key, and remain reachable for service messages. Singh expects current surfaces to become overcrowded and break under the number of agents that will use them. His advice is to build bridges from today's systems while learning from real use cases.
Agents want identity on surfaces people already use
Singh says agents want to live on the same surfaces as humans. Current products put agents in text, email, browsers, Slack, and coding environments, while humans provide their context, permissions, and tools. He argues that this matters because agents become useful through the systems people already understand and use. An email agent can interact with external parties, while a Slack bot can grow from a personal helper into a company brain. The surface gives the agent a place to operate and gives other parties a way to recognize and contact it.
Identity is accumulated through use and recognition
Singh rejects the idea that agent identity can simply be defined from scratch and issued to an agent. Human systems already judge identity through recognition, trust, history, failures, fallbacks, and accumulated behavior. He thinks agents will gain identity in the same way. An address, a record of messages, permissions, and repeated interactions can gradually establish who an agent is and what others should expect from it. This is why he sees existing systems as part of the path to agent identity, rather than temporary infrastructure that should be discarded as soon as a new standard appears.
Singh describes agent cards and passports as technically sound approaches that can state who an agent is, what it can do, and which permissions it has. He compares these designs to key cards that expire and can be revoked. Tighter scopes, faster expiry, and earlier revocation improve security, but Singh says they can also leave less room for agents to operate and gain adoption. His concern is that specifications focused on the final security model may not solve how agents join the existing services where people already work.
Singh uses MCP as an example of adoption through existing surfaces. One early example he saw spread was ordering DoorDash through Cursor. The protocol made tools available where people were already working, which gave agents useful capabilities without asking users to move to a new network. In his view, this turned familiar products into nodes for agent activity. A coding agent can reach many integrations, and an email agent can repeatedly handle tasks with external parties. The adoption comes from putting capability into a place people already use.
Email gives agents both trust and an attack surface
Email already has recognition, history, audit trails, DNS relationships, receiving rules, and controls such as SPF, DKIM, DMARC, and spam filtering. Singh says these properties make it a useful trust surface for agents. They also bring risks, including phishing, prompt injection, and outbound spam. He does not think the risks can be separated from the trust. Using an established surface means accepting its weaknesses as well as its accumulated recognition. Replacing it with a new system may remove those risks, but it would also remove the history and adoption that make the surface useful.
Telnyx used an inbox to complete autonomous signup
Singh describes a Telnyx phone carrier signup flow built for agents. The flow was designed for humans and assumed that the user already had an email address. AgentMail added an API call that could provide an inbox when an agent needed one, rather than forcing the agent to invent an address or fail during registration. The agent could then fill out the form, receive a verification code, get an API key, and keep the inbox as a permanent contact point. The inbox connected identity with both signup and later communication.
A permanent inbox makes the agent the service contact
After signup, the agent became the focal point for Telnyx communication. Telnyx could contact it when the balance was low, an account was suspended, support was needed, or a failure or recovery occurred. Singh extends the same idea to purchases made by agents. A transaction may require an email address for checkout, and later refunds, disputes, receipts, and audit questions require a record of who signed up and what it did. An identity surface therefore needs to support the whole relationship with a service, not only the initial authentication step.
Current surfaces will face overcrowding and larger blast radii
Singh says agents are moving from copilots toward coworkers with their own identities, tools, context, and permissions. They can join networks that already contain human trust and reputation instead of creating a separate network. He also expects this arrangement to break under stress. Existing surfaces may become overcrowded as agent numbers grow, and giving agents access to personal email or WhatsApp can send the wrong messages or expose more of a user's permissions. The resulting blast radius grows when one agent can act across many connected services.
