Hark Pro goes wide while Khosla-backed Wajo challenges Muse and Dot on trust

Hark Pro gets a wide release

Hark, a startup founded under a year ago by serial founder Brett Adcock, has moved its AI personal assistant Hark Pro into wide release, TechCrunch reports. The service is free to access, with a subscription tier for heavier users.

Hark describes its mission as building a user interface for AI rather than chasing AGI, and its underlying model is trained specifically for computer use — in effect, Hark Pro operates your computer on your behalf. The product is a full-screen experience built around a central chat window, a feed of suggested actions, and small dashboards the company calls "panels," set against background imagery matched to the user's local weather and time of day. Once granted access to email, calendar, files and payment cards, it can field ordinary chatbot questions or carry out digital chores.

In a demo for TechCrunch, design lead Abidur Chowdhury — a former Apple staffer — showed the assistant proactively flagging expenses to approve, emails to answer and a flight to book. He also recounted an instance where it noticed a colleague's car registration was due and completed the renewal through the California DMV website. One distinctive interface choice is that the agent's web navigation is displayed in a small window, something most competitors keep hidden, which Chowdhury framed as a trust-building measure. On positioning, he argued Hark Pro "will give you a better result" than rivals and "work end-to-end instead of asking you to take over," even if it is not faster than Meta's Muse. Hardware is promised for 2027 — an AI-native device whose form factor the company will not reveal — and TechCrunch notes an open question over how much general capability the specialized model trades away compared with a frontier LLM.

Wajo sells trust to wary users

The same day, TechCrunch profiled Wajo, whose Fo agent was built by founder Shivani Poddar, an engineer with stints at Meta, Google and DeepMind, including work on Facebook's first assistant and on bringing Gemini to market. Vinod Khosla is backing the startup, and his argument is blunt: because agents act on your behalf, trust decides the market, and Wajo is architected around trust and safety in a way rivals are not. He said he would not hand his data to Meta, since an ad-funded company makes the user "the product."

Functionally, Fo resembles Muse or Instinct: it works over iMessage and WhatsApp, presents task cards on the web under categories such as scheduling, gifting and life admin, and can phone businesses or individuals to book or cancel services. TechCrunch's early tests were mixed — the agent successfully called an airline about upgrades, but a personal call asking someone to feed the cats went less smoothly, and its disclosures were unclear. Fo can also hire a human to finish a task when needed, issue a virtual credit card that pays without exposing card details (a patent is pending), and use its own email address to correspond with people and businesses. When sites block agents, it can ask the user to step in or set up a group with the business for verification.

Wajo has also drawn backing from Jeff Dean and Gokul Rajaram, though the amount raised is undisclosed. The company says it is used in 107 countries and has grown tenfold since launching in late September, without sharing user numbers. On safety, it plans a "trusted contact" list governing which agents may call a given person, records requests not to be called again, and claims its architecture complies with Amazon's terms of service even as Muse has reportedly been blocked there. Multi-user social scenarios are next on its roadmap.

Why it matters

Meta's Muse, OpenAI's Dot and Instinct currently lead a category that is gaining new entrants weekly, and both newcomers are converging on the same thesis: the bottleneck is not raw model capability but users' willingness to hand over email, payments and calendars. Privacy-first positioning against the largest labs is becoming the standard wedge for startups, while norms for disclosure when an AI calls a human remain unsettled. Whether trust is a feature or the whole product will be tested over the coming year — and, in Hark's case, by a promised 2027 device that the company insists will be unlike anything built before.