Spark Capital
- Nabeel Hyatt of Spark Capital discusses AI's new reliance on diverse real-world data sources.
- Instawork is evolving by integrating robotics, affecting gig work platforms and job creation.
- Robotic data needs are creating new jobs in the gig economy.
Venture capitalist Nabeel Hyatt has spent his career backing companies that reshape how people interact with technology.
As a general partner at Spark Capital, he was an early investor in Discord and Cruise, two companies that redefined online communities and autonomous driving, respectively. He's also founded and led tech companies. Back in 2010, Zynga acquired Conduit Labs, where he served as CEO.
These days, Hyatt sits on the board of Instawork, a gig-work platform undergoing a dramatic evolution. The company, led by CEO Sumir Meghani, recently launched a robotics division and is building a new data-collection device, Instacore — an early attempt to solve what some researchers call the "100,000-year problem" of gathering enough training data for robots.
That vantage point gives Hyatt a front-row seat to one of the most consequential shifts in AI: the move from software-based models trained on internet data to physical AI systems that must learn from the real world.
Here's my interview with Hyatt about why robotics needs new kinds of data, and how gig workers may become central to the AI economy. It's been edited lightly for length and clarity.
Business Insider: Why do researchers and companies in robotics need so much new diversity and detailed data?
Nabeel Hyatt: Every robotics company is trying to train models that work in the real world, not just in a lab. And the real world is messy. Kitchens don't all look the same. Warehouses have different layouts. The way someone chops an onion in a hotel kitchen in New York is different from a catering facility in Houston.
There's no standardization on what "good" video data even looks like yet. OpenAI wants one thing, Figure wants another, Boston Dynamics cares about light industrial environments specifically. Some want egocentric video from a camera on someone's head. Others want teleoperation data. Others want 3D mapping of physical spaces. It's early enough that the whole field is still figuring out what data matters, which means you need massive variety: different environments, different tasks, different angles, different hardware.
So what you need is a singular company that has been out there in the real world, thinking about the human element for years, but the technical chops to be able to put that together. Instawork is uniquely positioned here, much like Handshake has been for us in that market.
How do you see the mid-to-long-term impact on gig work platforms? Will human labor be replaced by robots, or will new jobs emerge?
Both, honestly. But probably not the way people imagine. The near-term reality is that robots create jobs for humans. Someone has to go map a restaurant before a delivery bot can operate in it. Someone has to train staff on how to work alongside the robot. Someone has to do repairs when a kitchen robot breaks down. Longer term, yes, some jobs will be automated, but the pattern is jobs shift. They don't disappear.
Cars replaced the horse and buggy, and yes, that person taking care of the horse lost a job. But accessibility of the new technology meant that there are way more car mechanics today than there ever were people taking care of horses.
When did you first hear about the robotics push at Instawork?
Sumir had been talking about it for much of 2025, but it really started coming together around the fall. By the November board meeting, they had a conference room rigged up with cameras and wires, and engineers driving down to Santa Clara to upload terabytes of data into customers' systems. Aaron Bromberg (Instawork's head of robotics and applied AI) was joining the team, and you could feel the energy shifting. The demand was suddenly real, not theoretical.
Instawork is evolving from software to partly hardware. Are you concerned?
Not really. I mean, look, this isn't a company that woke up one morning and decided to become a hardware company. They're a labor marketplace with millions of workers, and it turns out the robotics industry needs exactly what they have: humans who can collect data in real environments. The hardware piece, the data capture devices, that came out of necessity. Nobody had the labor infrastructure to quickly deploy what Instawork can.
What's the ultimate goal for Instawork? 20 human pros going to a job plus five robots?
Something like that, yeah. Sumir describes it as two business lines: data collection is about 80% of the focus right now, and field ops (robot wrangling, servicing, deployment) is about 20%. But over time, those probably converge.
The vision is that Instawork is the platform you go to when you need humans who know how to work with robots. You're a restaurant chain deploying robots? Instawork sends you someone who's already certified on that system. You're a warehouse deploying units? Instawork has pros who've been collecting data in light industrial environments for years and know how to work alongside that hardware.
It's not "20 humans plus five robots." It's more like humans and robots become a blended workforce, and Instawork is the company that manages both sides of that equation. The company's real advantage isn't the technology; it's that they have the relationship with millions of workers and the operational muscle to deploy them into any new category of work that shows up, including categories that didn't exist two years ago.
Sign up for BI's Tech Memo newsletter here. Reach out to me via email at abarr@businessinsider.com.