Pronto’s Transition to Real-World Data Layer for Physical AI

Pronto is shifting from merely being a home service platform. Amidst the rapid growth narrative and excitement from investors, the startup is potentially targeting something much larger and possibly contentious: establishing a comprehensive real-world data foundation for Physical AI and robotics. This is not simply conjecture among industry experts. A key indicator arises from investor documents reviewed by Startup Superb.

In an internal communication, Glade Brook Capital mentioned: “Pronto aims to systematise India’s extensive informal labour markets, thereby generating data valuable for training physical AI and robotics.” The memo elaborates that the company is currently “testing real-world training data with top physical AI laboratories.” This insight fundamentally shifts how the company should be perceived.

Until now, Pronto was primarily viewed as a rapidly expanding instant home services startup that connects households with professionals for tasks such as cleaning, laundry, utensil washing, culinary assistance, gardening, and car detailing. Established in 2025 by Anjali Sardana, the company has rapidly grown across the ten largest cities and has garnered significant interest from investors. However, the Glade Brook memorandum implies that Pronto’s aspirations go well beyond home services.

According to the investor document, the company is “cultivating a data business that utilises its workforce to acquire real-world household data for robotics laboratories.” The memo also highlighted that initial partnership interest is “promising” and the startup is “acting swiftly to commercialise this strategy.”

In response to inquiries from Startup Superb, Pronto acknowledged that it has been conducting a limited pilot programme focusing on AI-related data initiatives. The company stated that it has contemplated this direction since its inception, adding that customers can voluntarily opt in to have their jobs recorded. Pronto explained that the professional uses “a small camera that faces outward at the work,” and customers receive the recorded material afterwards.

The startup further confirmed that Physical AI systems necessitate real-world behavioural data. “One type of data required is first-person video of individuals performing real tasks, like washing dishes and folding laundry, in genuine environments,” Pronto stated, emphasising that the work done by its professionals “can create a foundational data layer for physical AI.”

The consequences of this are substantial. Unlike online platforms that gather browsing behaviours or clicks, Pronto operates within one of the most private spaces imaginable: people’s homes. If household workflows are employed as training data for robotics systems, daily domestic undertakings could evolve into invaluable AI datasets. The ways in which individuals clean kitchens, fold garments, organise living areas, wash utensils, manipulate objects, navigate clutter, or operate appliances may all serve as critical inputs for Physical AI systems. This is precisely why investors globally are becoming more interested in this field.

Unlike generative AI, Physical AI demands real-world behavioural training. Robots require exposure to actual environments and repetitive human activities to learn effective operation within homes and workplaces. While synthetic simulations can assist, genuine data is far more beneficial.

The Glade Brook memo appears to acknowledge this opportunity directly. It indicates that Pronto is currently processing over 25,000 orders each day while piloting “real-world training data” initiatives with Physical AI laboratories. Sources familiar with the situation revealed that the company has assessed employing mounted cameras or workflow observation systems in select operational contexts.

Pronto has confirmed that it is using cameras as part of a limited pilot associated with AI initiatives, though it denied any extensive deployment. The company clarified that participation in the pilot is “strictly opt-in and selected by the customer at the time of booking, job by job,” emphasising that “by default, no one is included” and that it has “no plans to broaden this to the majority of customers.” When inquired about the pilot’s terms and conditions, Pronto did not disclose specific details and indicated that the initiative would be confined to select customers.

Under India’s Digital Personal Data Protection Act, 2023, consent must be both specific and purpose-bound. Agreeing to recordings for monitoring is legally different from consenting to their utilisation as AI training data. While Pronto presents the recordings as a customer service element, the investor memo categorises them as potential AI training content, raising issues regarding compliance with the DPDP Act’s purpose limitation criteria.

The company further stated that faces and identifying information are automatically blurred, no personally identifiable information is uploaded or shared, and footage is discarded within 48 hours. However, Pronto’s statements about data deletion lead to further queries. The firm claims that footage is erased within 48 hours and cannot be accessed by anyone other than the customer. Yet, both Pronto and Glade Brook Capital independently describe the initiative as generating real-world training data for Physical AI laboratories.

These two statements appear challenging to align, since AI training data usually necessitates curation, annotation, processing, and structured storage before it can be commercialised or used in training robotic systems. If household workflow data is indeed being documented or analysed for robotics purposes, an obvious question emerges: do customers and workers fully grasp what they are consenting to?

Could household layouts and behavioural routines potentially evolve into commercial AI assets? Who retains ownership of such data? Is it allowable to share it with third-party robotics companies or labs? Would users fully comprehend how Physical AI systems are trained? These inquiries gain urgency given that homes reveal far more than typical internet behaviour. Domestic settings can disclose deeply personal lifestyle, cultural, financial, and behavioural indicators. Even anonymised household datasets can retain highly sensitive contextual information.

Timing also plays a crucial role. Worldwide, investment is increasingly transitioning toward robotics, humanoid technologies, warehouse automation, and embodied intelligence following the generative AI surge. Investors believe that future AI leaders will be built not only on computational power and models but also on exclusive real-world datasets. This renders Pronto a strategically appealing entity.

The startup has successfully raised nearly $60 million to date, including investments from Glade Brook Capital and a recent $20 million contribution from Lachy Groom that has reportedly increased its valuation to around $200 million. While there is no evidence suggesting that Pronto is collecting data specifically for Lachy Groom’s interests in robotics or affiliated companies like Physical Intelligence, which aims to construct general-purpose AI models and learning algorithms for helping robots perform real tasks in various environments, the intersection of household workflow data, investor references to robotics training datasets, and growing investor enthusiasm for Physical AI is bound to raise eyebrows.

Queries directed to Groom and Glade Brook Capital on Thursday did not receive replies. Pronto articulated that its overarching goal is to enable informal workers to engage in the AI economy rather than be replaced by it.

“Professionals on Pronto can engage in the AI economy directly and benefit from the data their labour generates,” the company expressed. However, the broader implications extend beyond a single startup. If businesses operating within homes evolve into AI data infrastructure layers, India could ultimately become a leading supplier of real-world household behavioural data for international robotics systems. In such a scenario, differentiating between convenience platforms and AI infrastructure may become increasingly convoluted.