Hark, a startup founded less than a year ago by serial founder Brett Adcock, released its AI personal assistant Hark Pro to the public on Tuesday. The service is available free, with a subscription tier aimed at heavy users. The company positions the product as a user interface for AI rather than an attempt at artificial general intelligence, and it joins Muse, Instinct and Dots in the race to find the experience that makes an AI assistant a must-have.
Hark Pro is built on a model trained specifically for computer use, and it operates the user's computer on their behalf. It presents a full-screen experience with a homepage containing a central chat dialog and a feed of suggested actions and mini-dashboards the company calls panels. Its background is custom imagery generated for the user's local weather and time. As with rival assistants, users connect email, calendar, hard drive and credit card accounts so the tool can handle digital tasks alongside ordinary chatbot requests.
Abidur Chowdhury, a former Apple staffer who leads design at Hark, demoed the product and said it flagged expenses awaiting approval, emails needing replies, a meeting with a reporter and a flight that needed booking. He acknowledged the flight example has become a cliché, and noted that users often struggle to identify tasks the assistant can actually complete. He said he was impressed when it recognized a colleague needed to renew a car registration at the California DMV and completed the transaction through the agency's website.
Hark shows users a small window of the agent navigating the web, a step most competitors skip, which Chowdhury said is meant to build trust. He said the company is not Meta or OpenAI and is not selling ads or stealing data, a pitch that matters because using the models fully requires sharing one's digital life. Asked about profiling, he said Hark's work to understand users is task-related and focused on taking work off their plate.
Chowdhury argued Hark Pro's distinction is that it substitutes for the computer end-to-end rather than handing control back mid-task, and he invited comparisons with rivals, saying it may not be faster than Muse but produces better results. It remains unclear how much capability Hark trades away by using a computer-use model instead of a frontier LLM. The software is only half the plan: Hark promises an AI-native device in 2027, though it has not described the form, and Chowdhury said only that it will be something that has never existed before.
More AI news from TechManNews.
