Voracious demand for robotics training data is transforming gig work

• Instawork is transforming gig work with a new robotics business.
• The company launched Instawork Robotics Lab on Thursday and a robotics certification program.
• Instacore, a wearable camera system, is designed to collect detailed robotics training data.

The race to build intelligent robots is reshaping an unlikely corner of the economy: gig work.

As demand for real-world training data explodes, Instawork, a platform best known for supplying hourly workers to hotels, warehouses and stadiums, is transforming itself into a hardware-enabled robotics business, offering a glimpse of how human labor could evolve alongside machines.

The pivot highlights a gnarly bottleneck in the next chapter of the AI boom. While chatbots and large language models were trained on free data from the internet, there's no equivalent for robots. So the industry is creating its own from scratch, and Instawork is turning gig workers into real-world data generators to feed this voracious demand.

It started about a year ago when CEO Sumir Meghani checked on an Instawork shift that was held up. A group of Pros, as Instawork calls the gig workers on its platform, were stuck outside a warehouse in San Francisco's Mission District. The CEO dug a little deeper and discovered that a robotics company had posted the shift on Instawork's marketplace.

"I was like, 'why does this robotics lab need our people?'" Meghani told me during a recent interview at Instawork's headquarters. "Then, we started seeing more of these robotics shifts posted."

The tasks range from the mundane to the downright strange. Some shifts just require Instawork Pros to move a few robots here and there. Other shifts involve Pros operating robots from remote locations, and even recording themselves while working on tasks such as washing dishes or cleaning a room.

Meghani and his team started calling more of these robotics companies and soon realized there was a huge, new market that Instawork's platform could serve.

"With most great things that happen in business, it kind of happens organically," said Meghani, who recently renamed himself Chief Robot Officer. "Then you go chase it. "

From Instawork to Instacore

On Thursday, the first real fruits of this transformation emerged as the company announced the Instawork Robotics Lab alongside a robotics certification program that has already reached more than 20,000 Pros in its first few weeks. The program has two tracks: robotics and AI data collection, and certified robot technicians who can handle hardware resets, maintenance, and field support in more than 100 markets.

In May, the company plans to release Instacore, a wearable camera system designed in-house to help gig workers record and label their activities in more detail.

Instacore will collect this information and sell it to robotics companies and researchers, helping to train the physical AI models needed to produce robots that can truly understand and operate in the messy, real world.

100,000 year problem

The lack of robot training data is known as the 100,000-year problem, an idea popularized by Ken Goldberg, a UC Berkeley professor and cofounder of Ambi Robotics. Last year, he estimated that at current data-collection rates, a general-­purpose robot, based on a ChatGPT-­sized set of robotics data, will be available in 100,000 years.

"How can we close this 100,000-­year 'data gap'?" he wrote.

The tech industry is responding to the challenge because if robots can be trained to operate well in complex situations, the payoff could be huge. Goldman Sachs forecasts the humanoid robot market could reach $38 billion by 2035, and the global data-collection and labeling market is projected to hit $17 billion by 2030.

Instawork estimates the industry collected roughly 100,000 hours of robotic training data in 2024 and 1 million hours last year. In 2026, it's projected to hit 20 million hours. That's still only 0.04% toward closing the data gap, according to Meghani.

"Humans being humans"

All this has created what Instawork sees as a potent new market: one where skilled human labor is not displaced by robots, but monetized by helping to train the machines, deploy them, keep them running, and ultimately work alongside them.

Instawork, which has raised more than $150 million from investors including Benchmark, Greylock, and Spark Capital, is careful not to name customers. But it says it is already working with leading robotics labs and manufacturers.

The startup says those partners want two main things: better training data and a field workforce that can support robots after deployment. In that sense, Instawork's robotics push is both a data business and a labor business, with software, hardware, and services increasingly fused together.

Instawork plans to send thousands of certified Pros into real kitchens, warehouses, hotels, and other workplaces wearing Instacore, recording how they move, what they touch, and how they adapt when environments change.

The company says the data will be opt-in and anonymized, capturing workflows rather than identities. It is designed to be light enough for a full workday and flexible enough not to interrupt the job itself.

That balance matters because the value of the data depends on workers behaving like humans, not like subjects trapped inside a clunky research rig.

"We don't want to record humans simulating robots. We want humans being humans," said Instawork's robotics lead Aaron Bromberg, a former Amazon technologist who worked on that company's Astro home robot.

"Version zero"

I got a glimpse of this hardware-enabled future at Instawork headquarters on April 10. When I arrived, the place looked like a typical Silicon Valley software office, with a ping pong table, fancy kitchen, and rows of desks with slick computers and programmers tapping away.

I sat in a conference room waiting. Then, Bromberg walked in wearing four cameras strapped to his body along with a small backpack. He called this Instacore "version zero," and said a more streamline "version one" device is scheduled to launch in May.

Bromberg didn't let me take photos of the prototype for competitive reasons. But he walked me through each component in a detailed demo.

• Head camera: A head-mounted camera captures first-person "egocentric" video of tasks, with future versions adding stereo vision to measure depth and distance more precisely.
• Chest camera: A wide-angle chest camera records the broader environment, helping robots learn how tasks vary across real-world settings like different kitchens or lighting conditions. This is crucial because robots sometimes "freeze" in new environments, even if they've been trained to do tasks correctly, Bromberg said.
• Wrist cameras: Two wrist-mounted cameras capture detailed hand movements and finger dexterity, generating fine-grained data on how humans grip, manipulate, and handle objects. At one point, Bromberg grabbed a cup off a desk and showed me crisp, live video footage seen from the underside of his wrist. With that level of detail, Instacore can capture how hard humans are squeezing when they pick stuff up, incredibly valuable data for the AI models that power robots.
• IMU sensors: Each camera has built-in inertial measurement units to track motion, orientation, and acceleration, giving robots crucial spatial awareness data about how tasks are performed. For example, robots need to know to turn their wrists at a certain angle to pick up a bowl, and turn them even more to hold a tray.
• Backpack: A lightweight backpack houses a battery and on-device computing system, enabling all-day recording and local processing before data is brought back to Instawork.
• Instawork app: A companion mobile app guides workers through camera setup and calibration, ensuring consistent, high-quality data capture.

Two circles

The hardware matters because off-the-shelf options were not enough, Bromberg said. Full-body suits and bulky data-capture gear may be ideal for hardcore robotics research, but they are impractical for hourly workers who still need to complete real shifts.

Instawork is uniquely placed to develop a system that meets both needs, Bromberg told me as the Instacore chest camera bobbed below his chin.

With about 10 million Pros on its gig work platform, the startup has access to thousands of different tasks performed by hundreds of thousands of people. That provides the data quality and diversity that the robotics industry yearns.

Meanwhile, the clients hiring Instawork Pros still want tasks completely efficiently, so Instacore must be practical enough to use easily. These constraints produce a system that can operate at scale in most work environments while still collecting detailed data.

Bromberg stopped at this point and drew two circles on the conference room whiteboard. Over one circle, he wrote "robotics" and on the other "human." Where the circles overlapped, he wrote "Instawork."

"Crazy" board meeting

This is a striking evolution for Instawork. The company has historically been known as a software-based marketplace, not a hardware company. Hardware introduces more complexity, more capital needs and more operational risk.

"This is a big strategy bet for us," Bromberg told me.

Bromberg wore an early Instacore device to an Instawork board meeting, part of a pitch to persuade directors and investors that veering into hardware was worth the risk.

Meghani, the CEO, acknowledged the board had reactions ranging from skepticism to enthusiasm when he first presented the move.

"At some point, when we had to spend a bunch of money on hardware, I had to go to the board and let them know," he said. "The board's reaction was probably half 'you're crazy' and half 'This is amazing.'"

Bill Gurley, a leading venture capitalist, was at the board meeting and remembers the pitch. He's unruffled by the strategic pivot. While there's been some investment in data-collection and training equipment, Gurley said this is not a core part of Instawork's future gross profit margins.

That future profitability will be driven, in part, by long-term demand for robotics data and the emergence of new robotics jobs on Instawork's platform, he added. 

"It's impossible to automate without precision data. The real world is messy, not precise. So the training data needs to be a massive super-set," Gurley told me. "It will be a long time before we reach some form of saturation."

Three acts

Instawork's longer-term bet is that physical AI will create a new layer of work rather than simply replace old jobs. The startup sees its Pros not just as shift workers, but as trainers, technicians, and "robot wranglers" in an economy where machines increasingly need human expertise and support to function in the field.

Meghani, the CEO, describes the vision in three acts. Act one is the training data collected by Instawork Pros and the Instacore system. Act two involves supporting robots in the field, creating new "robot wranglers" who can repair, maintain, and troubleshoot machines.

In the final act, Meghani envisions Instawork's marketplace connecting companies with a hybrid workforce that combines human and robotic labor. Right now, Instawork sends 50 human Pros to staff an event at a hotel. In a few years, it might send 30 humans and 10 robots.

If the company is right, the next era of hourly work will not be human versus machine. It will be humans teaching machines how to work in the real world, showing up again when the machines need help, and ultimately working alongside the bots.

"There will be an essential role for skilled professionals in every chapter of this story," Meghani said. "We want to make sure our Pros are there for all of it."

Sign up for BI's Tech Memo newsletter here. Reach out to me via email at abarr@businessinsider.com.