Andreessen Horowitz Leads Conway Research’s First Funding Round

Andreessen Horowitz announced on October 2, 2026 that it is leading the first round of funding for Conway Research, the creator of Underdog, a personal artificial-intelligence product built to run on users’ own devices. The firm did not disclose the round’s size or terms.

The announcement was authored by Chris Dixon, a general partner who founded and leads a16z crypto; Gabriel Vasquez, an investment partner who focuses on AI application investments; and Elizabeth Harkavy, a partner on the a16z crypto investment team. The investors wrote that the firm had previously backed Conway founder Sigil Wen and shares his conviction that “the AI that knows us best should work for us, with data we control.” The firm said it believes that as Underdog improves its models and brings them to more consumer devices, it can put powerful AI in the hands of billions of people.

According to the post, Wen came to San Francisco at 17, slept for a time at a WeWork, and skipped college to work on AI. He spent nearly a year in a hacker house alongside researchers and engineers who helped create DALL-E, Whisper, GPT-3, and Stable Diffusion. Andrej Karpathy lived there, Anthropic cofounder Ben Mann invited Wen to become one of the earliest testers of what became Claude, and David Holz showed him the first version of the model behind Midjourney, the firm said.

The investors wrote that they first met Wen when Naval Ravikant brought him to an event Dixon hosted in 2021. Wen went on to help Ravikant launch Airchat, recruiting most of its engineering team, and Ravikant made him the youngest member of Spearhead. In 2025, Wen received the Thiel Fellowship.

Both the a16z post and Wen’s own account tie the product’s origin to an email breach. In a founder’s note published on the Underdog site, Wen states that when he was 18, a popular email app stored his emails on its servers and a bug leaked his private, privileged information, an episode he says cost him $100,000. The a16z post says the fallout helped motivate him to build products that keep user data safe. Wen also states that in 2023 he got GPT-2 running on his Apple Watch and that his team was first to ship an open speech model on the iPhone; a16z credits him with helping bring Whisper to the iPhone at Airchat.

The Underdog Product

Underdog is in an invite-only beta and is downloadable today for Macs with Apple silicon, according to its product site, which lists iPhone, iPad, Windows, Android, and Linux builds as coming. The company describes Underdog as a fully local AI model of less than 4 gigabytes that runs on the user’s computer rather than in a data center, and Wen states on his own page about the lab that the product’s models work without an internet connection.

The site’s listed capabilities include browser errands, such as booking flights, reserving tables, ordering groceries, and canceling subscriptions, with user approval required before any purchase or cancellation. An email client connects directly to Gmail and Outlook, with credentials stored on the user’s computer. Meeting notes are transcribed and summarized on the device, dictation works in any app with audio transcribed locally, and a calendar feature connects to Google Calendar and Outlook Calendar. The site also describes a concept-preview tool that locates a user’s personal information on data-broker sites and sends removal requests, and it lists expense tracking and tax preparation as coming soon.

On security, the company states that each account receives its own AES-256-GCM vault, with the encryption key held in Apple Keychain, secured by the Secure Enclave, and never leaving the device. Saved memories remain on the user’s Mac and can be inspected, edited, or deleted in Settings, according to the company.

Underdog is offered free. On its pricing page, the company states there is no server to pay for and no cloud AI reading user mail, leaving nothing to bill for, and it says the product will remain free because the user’s own computer does the work.

Conway Research and Its Inference Work

Conway Research describes itself on its website as a frontier AI lab building efficient models, inference engines, and personal AI products for devices people already own. Its stated thesis is “Underdog’s Law: today’s frontier intelligence reaches your devices in six months.” The lab lists nine research areas: agentic commerce, machine economies, programmable markets, autonomous organizations, robotics and autonomous retail, on-device AI, AI security, confidential inference, and megakernels.

On his page about the lab, Wen states that Conway also built Husky, an inference engine tailored to its Woof model on Apple silicon. He states that in the lab’s published benchmarks, Husky reached up to 4.5 times MLX’s token generation speed with Woof 4B on M5 Max, a peak recorded on a function-edit task with Flash enabled.

Wen also states that an essay he published in February 2026 titled Web 4.0 reached 10 million views in 48 hours. In it, he argues that most participants on the internet will soon be AI agents acting on behalf of a human, another agent, or entirely on their own, transacting at machine speed.