Khosla Says Trust Will Decide the Market for Personal AI Agents
Vinod Khosla says trust will decide the market for personal AI agents, and he is backing startup Wajo to win it.
In comments to TechCrunch published on Oct. 6, 2026, Khosla said "trust is the most important thing in an agent because it is acting for you." TechCrunch He described Wajo as more trustworthy than Muse or Instinct, saying it was built with trust and safety as the starting point.
Khosla tied that argument to data control. He said he did not want to give all his data to Facebook because Meta is in the ad business. His point was that large, ad-supported assistants should not automatically hold standing access to calendars, inboxes, purchase history and identity credentials.
Wajo was founded by Shivani Poddar, who previously held engineering roles at Meta, Google and DeepMind. Poddar helped build the first Facebook assistant. She told TechCrunch that "trust needs to be built from day zero in both architecture and product choices." The company has backing from Jeff Dean and Gokul Rajaram.
The product is Fo, Wajo's personal agent for errands. Fo works inside iMessage and WhatsApp, so users stay in existing message threads instead of opening a separate app for each request. It also has a web interface with cards of possible tasks. Wajo groups errands into categories including scheduling, gifting and life admin.
Fo is built to act across messages and payment systems. It can call businesses or other people to book or cancel services, and it has its own email address for talking with people and businesses. When software alone cannot finish a task, Fo can hire a person to finish it. That is human-in-the-loop, where a person handles physical-world jobs or unusual cases.
Payments are central to that approach. Fo can create a virtual credit card, a temporary number for one merchant or one use, and pay without sharing the user's real card details. The idea is to use narrow, single-purpose credentials that limit harm if misused, instead of passing around the main card number and billing identity.
Wajo has also built a set process for blocks and identity checks. When a website or service blocks AI agents, Fo tells the user what action to take or creates a group with the user and the business for verification. That keeps the user in the approval loop for CAPTCHA tests, login challenges, fraud checks and other anti-bot controls, instead of trying to bypass them quietly.
Wajo said Fo is being used in 107 countries and has grown 10x since its launch in late September. Khosla promoted Fo on Sept. 29, writing that "Fo is safer with credit cards" and that "Fo has much higher completion rates."
The broader context here is that agents change what access means. A chatbot that summarizes information only needs permission to read. An agent that books, cancels, pays and sends email needs permission to write, stored payment methods, a lasting identity and permission to speak as the user.
What is worth watching for practitioners is what Wajo is testing: whether people will give more authority to a narrower agent that limits payment data, shows blocks openly, and pulls a person into the thread when automation stops. If that holds, the edge goes less to the largest AI model and more to the system that best manages limited permissions, verification handoffs and visible user control. That would make scheduling and life admin as much an infrastructure task as a model task, with trust as the feature that allows more autonomy.
