Anjali Sardana, founder of Indian instant home-help startup Pronto, has established Provision Labs, a San Francisco-based venture focused on physical artificial intelligence (AI). According to a report by The Economic Times published on October 26, 2026, the new company 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 AI models for robotics and other physical applications.
Sardana is expected to serve as chief executive of both Pronto and Provision Labs, according to people familiar with the development cited by the publication. The move places her in a growing segment of the AI industry focused on teaching machines how to understand and interact with the physical world. While Pronto connects households with domestic-service workers, Provision Labs will focus on the data infrastructure that can help robotics developers train and evaluate physical AI systems. The company has not publicly disclosed detailed funding, customer or revenue information in the initial report.
Key Takeaways
- New venture: Pronto founder Anjali Sardana has established Provision Labs, a San Francisco-based physical AI company.
- Core focus: The company will collect data to train and test AI systems used in robotics and other real-world applications.
- Leadership: Sardana is expected to lead Provision Labs while continuing as Pronto’s chief executive.
- Industry opportunity: Robotics developers need real-world data to teach machines how to handle physical tasks and unpredictable environments.
- Connection to Pronto: The development follows Pronto’s exploration of using household-service activities to generate data for AI training.
- Unannounced details: Provision Labs’ funding, launch timeline, customers and commercial arrangements have not been publicly detailed in the initial report.
What Is Provision Labs?
Provision Labs is a new US-based physical AI venture established by Sardana, according to The Economic Times. The company plans to work with developers building AI models for robots and other systems that must perform tasks in the physical world.
Physical AI refers to artificial intelligence designed to perceive, understand and act in real-world environments. Unlike software that operates entirely on digital information, these systems must deal with objects, movement, spatial relationships and changing surroundings.
For example, a robot operating in a warehouse must recognize objects, understand where they are positioned and determine how to pick them up safely. A domestic robot would face additional challenges, including different room layouts, fragile objects and tasks that require adapting to human behavior.
Training such systems requires data that captures how the physical world behaves. Provision Labs intends to work in this area by collecting information that developers can use to train and evaluate their models.
The venture’s precise data-collection methods, technical infrastructure and initial customers have not yet been disclosed in the available reporting.
Why Physical AI Needs Real-World Data
Robotics companies have made progress in computer vision, machine learning and language-based interfaces, but reliable physical action remains a difficult problem.
A model trained on text and images may recognize a cup or understand an instruction to place it on a table. A robot must go further: it needs to estimate the cup’s position, determine how to grasp it, apply the right amount of force and respond if the object moves.
Real-world data can help developers train systems to handle these situations. Depending on the task, useful datasets may include video, camera images, sensor readings, movement trajectories and demonstrations of how people manipulate objects.
| Data type | How it can help robotics |
|---|---|
| First-person video | Shows tasks from the perspective of a person performing them |
| Images and spatial observations | Help models identify objects and understand their surroundings |
| Movement trajectories | Demonstrate how hands, tools or robotic components move during a task |
| Sensor data | Helps systems interpret distance, force, contact and other physical signals |
| Task demonstrations | Provide examples of how a task can be completed from beginning to end |
The commercial opportunity is to provide robotics developers with useful, well-organized training and evaluation data without requiring every AI lab to collect all the examples itself.
However, collecting more data does not automatically produce better robots. Data quality, diversity, accurate annotation and testing in real operating conditions all affect whether a model can perform reliably.
How Pronto Fits Into Sardana’s Physical AI Ambitions
Sardana launched Pronto in 2025 to connect households with workers providing everyday domestic services. The startup operates in India’s rapidly developing instant home-help market, where companies compete to make cleaning and other household tasks available on demand.
Pronto has also explored a connection between household services and AI training data. In a separate initiative called Verified, the company introduced an opt-in service involving cameras worn by participating workers during certain bookings. The footage was intended to support physical AI data collection, according to reporting by Mint.
That initiative generated discussion about privacy because recordings inside homes can capture sensitive information about residents and their surroundings. Pronto said participation was optional and customers needed to consent to the recording service. The company also described measures intended to protect privacy and process footage.
Provision Labs’ reported focus on collecting data for robotics developers is consistent with the wider commercial interest in physical AI datasets. However, the available announcement does not establish exactly how the new venture will obtain its data, whether it will use Pronto-generated datasets, or whether the two companies will have a formal commercial relationship.
Those distinctions matter because household environments can offer valuable examples of real-world tasks, but collecting data in private spaces requires careful handling of consent, security and personal information.
Why Robotics Companies Need Data Infrastructure
AI development has traditionally relied heavily on digital information, including text, code and images. Robotics introduces additional challenges because a model’s decisions can have physical consequences.
A robot may encounter unfamiliar objects, unexpected obstacles or changes in lighting and surface conditions. It must also learn how its actions affect the environment. Training and evaluating those capabilities requires data that represents varied physical situations.
Dedicated data companies can help by collecting demonstrations, processing raw recordings, removing sensitive information where appropriate and organizing datasets for model development.
Provision Labs will enter an area attracting attention from robotics companies and AI infrastructure providers. Its opportunity will depend on whether it can supply data that is sufficiently diverse, reliable and relevant to developers’ needs.
The initial report does not identify Provision Labs’ customers or confirm any contracts. It is therefore too early to assess the company’s scale or commercial position.
What This Means for the Physical AI Industry
Sardana’s move reflects a broader effort to build businesses around the infrastructure required for robotics, rather than developing robots or foundation models alone.
The physical AI ecosystem includes model developers, robot manufacturers, simulation providers, data-collection businesses and companies that evaluate how systems perform in real environments. Each addresses a different part of the development process.
A specialized data provider could help robotics developers reduce the time and expense involved in collecting and preparing training material. It could also make it easier for smaller teams to access data that would otherwise require extensive field operations.
At the same time, physical AI data presents challenges that are less pronounced in purely digital applications. Recordings may contain identifiable people, private homes or proprietary industrial processes. Businesses must establish appropriate consent, security and data-handling practices, especially when collecting information from workers or customers.
The long-term value of Provision Labs will depend not only on how much data it can collect, but also on its usefulness for training robots and the safeguards surrounding its collection.
The Bigger Picture
Provision Labs’ creation adds another development to Sardana’s work at the intersection of physical services and artificial intelligence. Pronto addresses the organization and delivery of domestic labor, while the new venture is intended to support developers building AI systems that can perform physical tasks. Whether the two businesses share data or other resources remains unconfirmed.
More broadly, the robotics industry is developing an ecosystem around the data required to make machines useful outside controlled demonstrations. As robots are expected to operate in homes, warehouses, factories and other complex environments, demand for realistic training and evaluation material could grow. Companies that can collect high-quality data responsibly may become important suppliers to this emerging market.
FAQs
1. What is Provision Labs?
Provision Labs is a San Francisco-based physical AI venture established by Pronto founder Anjali Sardana. It will focus on collecting data to train and test AI systems used in robotics and other real-world applications.
2. Who is Anjali Sardana?
Anjali Sardana is the founder and chief executive of Pronto, an Indian instant home-help startup. According to The Economic Times, she is also expected to serve as chief executive of Provision Labs.
3. What is physical AI?
Physical AI refers to AI systems designed to perceive and act in the real world. Examples include robots that handle objects, navigate spaces or carry out tasks in homes, warehouses and factories.
4. How will Provision Labs help robotics companies?
The venture plans to collect data that developers can use to train and test AI models for robots and other physical applications. Its specific products, customers and data-collection methods have not yet been fully disclosed.
5. Will Provision Labs use data collected through Pronto?
The available report does not confirm a formal data-sharing arrangement between the two companies. Pronto has separately explored opt-in household-service recordings for physical AI training, but any connection to Provision Labs remains unconfirmed.
Looking Ahead
The next developments to watch are Provision Labs’ initial customer relationships, the types of datasets it intends to provide and whether it announces funding or a commercial launch. These details will help clarify how the company plans to compete in the growing market for physical AI infrastructure.
For Sardana, leading two ventures across home services and physical AI data will bring distinct operational demands. Provision Labs’ prospects will depend on its ability to serve robotics developers with useful data while addressing the practical challenges of quality, privacy and responsible collection. The company’s early execution should provide a clearer indication of its role in the broader robotics ecosystem.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
