Adi Singh

Adi Singh co-founded AgentMail, which gives AI agents persistent email inboxes for two-way communication and service registration. His central argument is that agents develop an identity through repeated interactions on channels people already use: an address others recognize, a conversation history, and a way to remain reachable after a task ends.

Building an inbox for agents

Before AgentMail, Adi held investing roles at Accel, StepStone Group, and Flex Capital, alongside operating experience in accounting, education technology, and e-commerce. He co-founded AgentMail with Haakam Aujla and Michael Kim; all three are University of Michigan graduates. The company joined Y Combinator’s Summer 2025 batch and announced a $6 million seed round led by General Catalyst in March 2026.

The team’s infrastructure grew out of building email agents themselves. Existing providers made programmatic inbox creation difficult and imposed account structures and limits designed for human users. AgentMail made the persistent inbox the product: an address an agent could receive messages at, reply from, and return to after its current session ended. Threaded replies and incoming-message events support ongoing exchanges rather than isolated outbound notifications.

Adi initially expected developers to remain the immediate users of that infrastructure. By April 2026, he was reconsidering that assumption: agents were using the service to register for accounts, contact people, and establish their own points of contact. That shift raised a concrete question of responsibility: who owns an agent’s address, and who can limit what it does through that address?

Identity through use, permissions through infrastructure

Adi also revised an early expectation that agent-to-agent protocols would quickly become central to AgentMail. He argues that new protocols lacked consistent uses while agents gained adoption inside tools people already used. He distinguishes a credential that grants temporary access from an identity built through recognition, history, and reputation. An invoice-processing agent’s address, for example, becomes the place vendors know to contact it.

The Telnyx signup example he discusses makes this argument concrete. An AgentMail inbox lets an agent verify itself, obtain an API key, and remain reachable for billing and support. The inbox serves both the initial transaction and the continuing relationship. Adi’s emphasis is on making agents useful within existing relationships before designing an entirely new network around them.

Email also gives strangers a way to place instructions in an agent’s context. In his writing on agent email security, Adi describes the risk of a persuasive message asking an agent with file access to forward a private document. Authenticating the sender does not establish that the request is safe: a correctly authenticated email can still be malicious.

He advocates restrictions enforced outside the model: allowed senders and recipients, separate customer resources, and API keys limited to particular inboxes or actions. These controls constrain what an agent can do even when an incoming message persuades it to act. They are capabilities of the team’s infrastructure, rather than evidence that Adi personally implemented every component.