Pronto Founder Anjali Sardana Sets Up Physical AI Venture Provision Labs In US

The launch follows scrutiny of Pronto's household recording pilot, which the startup said was strictly opt-in and covered less than 0.01% of its users

Instant home services startup Pronto's founder Anjali Sardana has launched Provision Labs, a physical AI venture based in San Francisco.

According to its website, Provision Labs will collect data to train and test AI systems designed to perform real-world tasks using robots and other machines. It will work with developers building physical AI models.

The development was first reported by Economic Times. Provision Labs plans to source training data from professionals across sectors globally and help AI developers deploy their models, the report said.

Sardana will serve as the CEO of Provision Labs as well as Pronto.

The launch comes months after Pronto faced scrutiny over a pilot involving recordings of household chores to generate training data for physical AI and robotics systems.

The issue came to light after Entrackr reported that Pronto was exploring the use of footage from household tasks to generate real-world training data for robotics labs. Investor documents reviewed by the publication stated that the startup was building a data business by leveraging its workforce to capture household activities and had begun piloting real-world training data with leading physical AI labs.

During the pilot, Pronto's service professionals wore cameras while doing household chores such as washing dishes, folding laundry, cleaning and preparing meals.

Pronto later told Inc42 that the pilot was voluntary and only available to customers who opted in while booking a service. The startup said the pilot covered less than 0.01% of its users at the time.

"Pronto does not, by default, record in customers' homes, nor does it use customer data. We started the pilot you are referring to a few days ago. It is a strictly opt-in feature, available only to customers who actively choose it at the time of booking," the startup said back then.

Pronto also said participating customers paid for the feature and could access anonymised footage on the app for 48 hours. It maintained that the footage would not be shared with third-party AI labs and would be deleted from its servers within 48 hours.

In a subsequent social media post, the startup said it had spent months working towards compliance with India's Digital Personal Data Protection (DPDP) Act. It argued that real-world behavioural data would be essential to training robots to operate in household environments.

However, the pilot drew criticism over whether customers fully understood how recordings made inside their homes could be used to develop physical AI systems. Questions also emerged around data storage, processing and the extent to which such footage could be used to build commercial training datasets.

Physical AI Creates Demand For Real-World Training Data

Provision Labs' launch comes amid growing demand for real-world data to train robots to manipulate objects, navigate different environments, and perform everyday tasks.

Startups are already tapping India's workforce to build these datasets. Y Combinator-backed Human Archive collects recordings of workers performing physical tasks, processes the footage, and sells the resulting datasets to frontier AI labs and robotics companies.

The likes of Awign, Objectways, and Human Archive are building pipelines to supply first-person footage of human activity to AI developers. However, the opportunity raises questions around workers' consent and compensation, the privacy of others captured on camera, and how much value Indian data suppliers retain when the resulting models and intellectual property belong to overseas companies.