Instawork turns its gig marketplace into a robot-training data line
Instacore puts five cameras and a compute backpack on workers to capture commercial tasks for AI labs, with customers still unnamed.
By Ryan Merket ยท Published
Why it matters
Instawork is testing whether labor marketplaces can become data suppliers for physical AI. If robotics labs need real-world task data more than another simulation run, platforms with worker networks and commercial-site access gain new leverage.
Instawork CEO Sumir Meghani is turning the company's hourly labor marketplace into a physical AI data operation, launching Instacore, a wearable camera system for workers to record commercial tasks that robotics companies can use to train models, Business Insider reported on June 9.
The move did not start as a pure robotics strategy. Meghani told Business Insider that Instawork noticed robotics companies posting shifts on its platform last year, then he began walking with robotics founders near Y Combinator's Dogpatch headquarters to understand the demand. "People start to use your product in ways you didn't intend, and then you follow that signal," Meghani told the publication. He described robotics as "the intersection of physical labor - which we are really good at - and AI."
That is the useful read on Instacore: not a gadget launch, but a marketplace company trying to monetize the physical work it already routes. Instawork is best known for matching hourly staff with shifts at hotels, warehouses, stadiums and similar venues. Business Insider says Instawork has about 10 million workers, whom the company calls Pros, on its platform. If that number reflects active supply at meaningful scale, it gives Instawork something robotics labs are short on: access to messy, ordinary work in places built for people rather than robots.
A labor marketplace becomes a sensor network
Instacore includes five cameras mounted on the head, chest and wrists, connected to a compute backpack built to last an eight-hour shift, according to Business Insider. The head cameras capture depth, the chest camera gives broader environmental context, and the wrist cameras track hand movement. Business Insider reported that the finished system weighs less than three pounds.
The intended tasks are the kind simulation does not fully capture: chopping vegetables in a commercial kitchen, stocking shelves in a grocery store, or performing warehouse and hospitality work under real lighting, clutter, interruptions and human motion. Instawork says the data collection will be opt-in and anonymized. Business Insider also noted that in California, New York and Illinois, workers on Instawork's platform are treated as employees in legal terms rather than gig workers.
Instawork has raised more than $150 million from investors including Benchmark, Greylock and Spark Capital, Business Insider reported. The article did not identify a new financing tied to Instacore, a valuation, or named customers. Instawork told Business Insider it is working with leading research labs, but declined to identify them.
The hard part is calibration, not the cameras
The public launch was accompanied by a technical thread from Ryan Hickman, who described the infrastructure behind what he called the "Instalab."
The thread said the operation uses automated calibration stations with robotic arms that sweep camera targets to resolve camera matrices, IMUs and wrist-to-tag spatial transforms. Hickman also wrote that the team verifies "sub-millisecond temporal alignment."
That detail matters because raw video is not enough for robotics training. If wrist motion, head pose, depth and environmental context are not aligned in time and space, the resulting dataset can be noisy in exactly the places a robot policy needs precision. The thread said the Instalab floor was empty in April and was converted within weeks by assembly Pros and "IRL Captains" into a hardware and computer-vision line that could adapt to calibration and QA flows.
Business Insider identified the Mountain View operation as a production buildout where Instawork is assembling hundreds of Instacore systems, with plans to move production overseas to scale. Parts are sourced from China and the US, and some components are 3-D printed in-house, according to the report.
Aaron Bromberg, Instawork's head of robotics, is the operator attached to the hardware problem. Business Insider said Bromberg previously worked at Amazon on Astro, the home robot. He told the publication that early Instacore versions were too heavy, too hot and did not have enough battery life. Those are mundane constraints, but they decide whether a worker can wear the system through a shift without turning data collection into a stunt.
Instawork is selling reach into the physical world
Instawork's bet lands as AI companies are moving from text and image models toward systems that act in physical environments. Business Insider framed Instacore alongside robotics efforts at OpenAI, NVIDIA (@nvidia), Meta (@Meta) and Tesla. RuntimeWire recently covered the still-unconfirmed claim that OpenAI's Sora team had become its robotics team, a reminder that the market is reading every AI lab reorganization through a physical AI lens.
The closest comparison in Business Insider's reporting is shift, which offered free apartment cleanings in New York if cleaners wore head-mounted cameras while doing chores. Instawork's advantage is different: it already has a marketplace connected to commercial kitchens, warehouses, stadiums, hotels and other work sites where robots would eventually need to operate.
The unresolved question is whether Instawork can turn that access into datasets robotics labs will pay for repeatedly. The company has a plausible supply-side wedge, an installed labor network and a production line that appears designed for calibration at scale. What remains undisclosed is the demand side: which labs are buying, under what terms, and whether workers will view recording the physical details of their jobs as a paid opportunity rather than surveillance with a backpack.