a16z leads Conway Research's first round for on-device personal AI

Why it matters

a16z is backing a founder whose personal privacy breach informs Conway's local-AI thesis. The financing supports the ambition to build personal intelligence for hardware people already own, along with the underlying model and inference work. Conway has disclosed no funding terms or product traction.

Andreessen Horowitz is leading Conway Research's first funding round, backing founder Sigil Wen as he builds Underdog, a personal AI product designed to run on users' own devices. The October 2nd announcement did not disclose the round's size or valuation. Andreessen Horowitz says it had backed Wen before, but did not specify the timing or structure of that earlier support.

For Wen, the product's local-first approach grew from a personal privacy failure. The a16z account says a bug in an email app exposed his private, privileged emails to other users while he was living and working alongside AI researcher Andrej Karpathy. The incident helped shape his view that an assistant should be able to use a person's sensitive information without requiring that information to leave devices the person controls.

Conway says it aims to build personal intelligence for hardware people already own. Its website describes Conway as a frontier AI lab working on efficient models, inference engines, and personal AI products, with the thesis that frontier intelligence can reach users' devices within six months.

A founder who learned by building alongside AI researchers

Wen's route into the field ran through work and relationships. According to a16z, he arrived in San Francisco at 17, spent time sleeping at a WeWork, skipped college, and lived for nearly a year in a hacker house with researchers and engineers connected to DALL-E, Whisper, GPT-3, and Stable Diffusion. Karpathy lived there; Anthropic co-founder Ben Mann invited Wen to test an early version of what became Claude, and Midjourney founder David Holz showed him an early version of the model behind Midjourney.

Wen later helped Naval Ravikant launch Airchat, where he brought emerging AI models into the product and recruited most of its engineering team, according to a16z. The SMA Immigration profile also describes Wen as a co-founder of Airchat and says he built its invite loop. In 2025, he received the Thiel Fellowship, the investment firm said.

Before Conway, Wen's work ranged across products and early-stage company building. SMA's profile describes Serendipity, a relationship-management product, and BitSwap, a project Wen built in high school. At Conway, a16z is financing Wen's next attempt to turn an idea about software and trust into a daily-use product.

The privacy promise has to work as a product

Underdog is Wen's answer to the problem of giving an AI useful personal context while keeping that context under the user's control. The a16z announcement says the product runs on users' devices and can help with decisions and time-consuming digital work, including offline. It frames the bet around a growing tension: assistants become more useful when they can draw on email, calendars, documents, and preferences, the same information users have strong reasons to protect.

Wen had explored the technical side of that bet before Conway. The investors say he got GPT-2 running on an Apple Watch and later helped bring Whisper to the iPhone at Airchat. Local AI depends on models and software that can do useful work within the memory, battery, and compute limits of everyday devices.

Conway's public materials also describe research into on-device AI, security, confidential inference, and efficient model execution. Conway presents these as part of a broader lab, not just a consumer app project. Making its on-device assistant useful depends on the underlying model and inference work.

Keeping processing local can make privacy and offline use central product features, while also limiting the need to send personal data to a provider's servers. Those advantages have to show up in a product people choose to use. The announcement provides no user, revenue, pricing, or retention figures, and it gives no terms for the round. The investment establishes a16z's support for Wen's thesis; it does not establish demand for Underdog or the economics of delivering it.

A16z says it believes the opportunity could eventually reach people through consumer devices, and argues that capable AI should give users more control over their time and information. That is an investor's forecast, not a reported customer milestone. Wen's experience gives the product a clear origin, and Conway has a founder with a record of building in AI's orbit. The test is whether users will rely on an assistant built around local execution and data control.