Pronto Founder Anjali Sardana Launches Physical AI Firm Provision Labs

Anjali Sardana, founder of the instant home services startup Pronto, has established a new San Francisco-based venture called Provision Labs. The company will focus on collecting data to train and test artificial intelligence systems designed to perform real-world tasks using robots and other machinery.

Expanding into Physical AI

According to the company’s website, Provision Labs intends to collaborate with developers building physical AI models. The venture plans to source training data from professionals across various global sectors to assist in deploying these models. Sardana will hold the position of CEO at both Provision Labs and Pronto.

The launch follows a period of scrutiny regarding Pronto’s previous efforts to generate training data. Reports indicated that the startup had explored using footage of household chores—such as washing dishes, folding laundry, and preparing meals—to train robotics systems. During a pilot program, service professionals wore cameras while performing these tasks in customer homes.

Pronto previously stated that the pilot was voluntary and limited to customers who opted in during the booking process. The company noted that the program affected less than 0.01% of its user base at the time. Pronto also maintained that participating customers could access anonymized footage for 48 hours, after which the data would be deleted from servers. The startup asserted that it does not record in homes by default and emphasized its efforts to comply with India’s Digital Personal Data Protection Act.

Industry Demand for Real-World Data

Despite the company’s assurances, the pilot faced criticism regarding data storage, processing, and whether customers fully understood the potential commercial use of the recordings. Critics questioned the extent to which footage from private homes could be utilized to develop commercial training datasets.

The creation of Provision Labs comes as the tech industry faces rising demand for real-world data. Developers require this information to train robots to manipulate objects, navigate diverse environments, and complete everyday chores effectively.