The race for the ultimate personal AI assistant gained two competitors this week that make privacy their main selling point. On Tuesday, October 6, 2026, the startup Hark broadly released Hark Pro. A day earlier, on Monday the 5th, young programmer Sigil Wen opened an invite-only beta of Underdog. TechCrunch covered both launches in separate articles and placed them in a market where products such as Muse, Instinct and Dots already compete.
Hark was founded less than a year ago by serial founder Brett Adcock. Its service can be used for free, with a subscription tier for heavy users. According to TechCrunch, the company does not present itself as a lab chasing artificial general intelligence or a cure for cancer, but as a builder of user interfaces for AI. Its foundation is a model trained specifically for computer use, and that is what Hark Pro does: it uses your computer for you.
The experience is full screen. Hark Pro displays a kind of homepage with a central chat box and a feed of suggested actions the user might need, plus small panels with data of interest. The background is a custom image tailored to the local weather and time. As with other assistants, the user grants it access to email, calendar, hard drive or credit cards so it can carry out digital tasks and answer questions.
Abidur Chowdhury, a former Apple designer who leads design at Hark, gave TechCrunch a demo. The assistant proactively asked him about expenses awaiting approval, unanswered emails and a meeting, and offered to book a flight. Chowdhury said he was impressed when the assistant noticed that a colleague needed to renew a car registration with the California DMV and completed the process on the agency's website.
One design choice stands out: Hark Pro shows in a small window how the agent navigates the web, something most assistants do without displaying it. According to Chowdhury, the goal is to build trust so that users can see the agent is doing the right thing. That transparency is part of the sales pitch. “We're not here to like sell you ads and steal your data,” he said, referring to companies such as Meta and OpenAI.
Chowdhury argued that Hark Pro aims to fully substitute for using a computer and that, in his view, it simply works better than its rivals. “It might not be faster than Muse, but it will give you a better result, and it will work end-to-end instead of asking you to take over at a point,” he said. TechCrunch raises a reasonable question: a model trained only for computer use is faster and cheaper at those tasks, but it is unclear how much it trades away compared with the broad knowledge of a frontier model.
The software is only half the bet. Hark promises to release an AI-native device in 2027 to support its software. The company has not revealed its form, though Chowdhury criticized the “creep factor” of wearable cameras. “The first product that we're building is a product that's never ever existed before,” he said. “It's so hilarious to me that you have AI now, you have a chatbot, and you type stuff in with a keyboard. Fifty years later and the future of computing is still a terminal.”
Underdog starts from a more radical idea about privacy: everything runs on the device. Its creator, Sigil Wen, moved to Silicon Valley at 17 and lived in an AI hacker house with researcher Andrej Karpathy, as he told TechCrunch. Today he is a Thiel Fellow, the program from investor Peter Thiel that encourages young founders to build companies instead of going to college. The startup behind the product is called Conway Research.
The assistant currently runs on Macs and Windows PCs, and Wen promises Linux, iPhone and Android versions soon. It uses Husky, an inference engine Wen built himself to run models quickly on the user's own hardware; according to him, it moves less data between the main processor and the graphics chip than other local engines. It also encrypts the keys to the email and other accounts the user authorizes.
The trade-off is model size. Underdog uses a 27-billion-parameter reasoning model fine-tuned from Qwen3.8-27B, far smaller than the large models hosted in data centers. Wen argues that it compares favorably with Claude Opus 4.6 on some benchmarks, which was considered top-tier performance six months ago, and that it is enough for everyday tasks such as shopping research or math homework questions.
The business model is the most original part. The app will be free at first and never ad-supported. Because the AI runs on users' machines, Underdog does not pay a provider for inference. Instead, with Stripe co-founder Patrick Collison among its investors, Wen plans to take a small percentage of the payments the assistant makes through Stripe's payment rails, similar to an interchange fee, without needing to mine user data.
Conway Research is backed by Andreessen Horowitz, through partner Chris Dixon, Khosla Ventures, Hummingbird, SV Angel and the Anthology Fund, the partnership fund between Menlo Ventures and Anthropic. Its angel investors include Vercel founder Guillermo Rauch, Noam Brown and Deedy Das. “I'm building a product that I would be proud for my future children to use,” Wen said.
Both launches respond to the same tension. To be useful, a personal assistant needs access to the most intimate parts of digital life: health, finances, family. TechCrunch notes that many privacy policies in the industry allow companies to collect data that may be sold to advertisers or used to train other models. Hark promises task-focused use; Underdog goes further and keeps the data from leaving the computer.
In conclusion, Hark Pro and Underdog show two different paths to the same problem: visible trust in the cloud, in Hark's case, and privacy by design on the device, in Underdog's. Neither has yet proven its performance against the giants, and their claims come from the founders themselves. But both turn privacy into a competitive argument in a market where assistants want to take over our most personal tasks.
