Inkbox Founder Ray Liao: YC Ends with $300K ARR
Inkbox founder Ray Liao stated that when entering Y Combinator, the revenue was zero, and on demo day, the annual recurring revenue crossed $300,000. He asked partner Tyler Bosmeny why he didn't return the call after the interview, to which Bosmeny replied that he had conducted product due diligence with Boris Jabes to confirm that the experience matched the description.
Inkbox is part of YC's Summer 2026 batch, positioned as an identity and communication layer for AI agents: providing agents with independent email, phone, iMessage, and internet addresses, and offering a delegation channel between agents to consolidate multi-channel conversations into a single context. The three-member team is based in San Francisco, with Liao serving as CEO, holding a PhD in computer science from MIT, and previously co-founding the cardiovascular virtual clinic Empallo; co-founders also include Dima Vremenko from Harvard Medical School with a machine learning background, and Alex Wilcox from Northeastern University with a background in computer science, focusing on machine learning and network research. The group partner is Bosmeny, co-founder of Clever. The product is compatible with Claude Code, Codex, Hermes, OpenClaw, etc., and provides SDKs and plugins.
The company was founded in 2025 and publicly launched the identity layer in July 2026; it had previously launched another product. Use cases include pairing engineering, growth, operations, and assistant agents with real email addresses and numbers. Communication between agents is designed as a mediator rather than a direct point-to-point connection. On S26 demo day, about 240 companies presented to around 1,500 investors, with $300,000 ARR being a notable revenue figure in a "season starting from zero," but still lower than the nearly million or higher reported by leading companies in the batch.
After the interview, Bosmeny first tried the product before returning the call, changing the accelerator's selection from pitch intuition to hands-on validation. The jump from zero to $300,000 occurred within the same batch window, corresponding to the need for agents to have a formal identity that can be dialed or emailed, rather than just a conversation in a terminal.
In market mechanics, this is a subscription for developer infrastructure, not consumer traffic. The buyers are teams looking to move agents out of chat windows and into email and phone networks; the sellers are three-person companies packaging numbers, emails, and tunnels. Beneficiaries are YC companies and toolchains that have already treated agents as colleagues; the pressured parties are assistant products that only provide models without reachable identities. Funding was repriced on demo day based on annual recurring revenue metrics, with investors buying a seat fee for the identity layer, not the model itself.
Supplementary structure: The team emphasizes credential vaults and secondary verification, as agents must receive emails and verification codes when logging into the portal. The phone stack previously matched high-priced communication interfaces with open-source solutions.
Source: Public information
ABAB AI Insight
Once agents can write code, what they lack is a business card that can be reached by the world. Inkbox is not selling a larger model, but rather email addresses, numbers, and iMessage handles. Liao leveraged his MIT PhD and experience from a previous medical AI company to convince Bosmeny to first test the product before deciding to call—this is the accelerator shifting from "sounding smart" to "can we send a text to the agent tonight." The jump from zero to $300,000 ARR compressed into one summer batch indicates that the willingness to pay for the identity layer has emerged, but the scale remains a tool subscription, not platform rent.
The capital path follows the YC demo day auction. Nearly 240 companies in S26 competed for the same batch of checks, with revenue figures being more concrete than architectural diagrams. Bosmeny and Jabes' product due diligence reduced the discount on the story of "three PhDs building infrastructure." The plugins first integrated with Claude Code and Codex, embedding distribution into existing agent workflows rather than retraining users to open a new client. Communication fees and seat fees generate cash flow, while model companies continue to collect tokens.
Analogous to Twilio making numbers programmable and early Auth0 turning login into a service: first occupy existing human channels, then allow new entities to move in. The industry phase is agents transitioning from terminal assistants to members of organizational contact lists. Whoever controls the agents' legal communication identity stands at the charging point for task delegation and customer reach. Point-to-point agent protocols are made into intermediaries to retain auditing and billing, rather than for philosophical purity.
Structural judgment belongs to the technical replacement of organizational interfaces. The mechanism is: models can perform tasks, but banks, hospitals, and suppliers still only recognize emails and phone numbers; the identity layer pushes non-human entities into old channels, thus transforming the measurement unit of old channels into ARR. Achieving $300,000 from zero in 12 weeks requires that the channel itself already exists; the startup merely facilitates the agents' network access.
ABAB News · Cognitive Laws
After agents can work, they first need a business card that can be reached.
If the interview call is not returned, it may be because the other party is testing your product.
For new entities to make money, they must first squeeze into the old world's email and address book.