When I first met Rajat Bhageria 8 years ago, he quietly let me know that he was also building a food robot company. And for the next half decade, that’s about all he told me.

The stealthiness continued as he quietly worked with his team to get his product to market and land some big customers.

Now, there are no more tight lips. In fact, after Bhageria and Chef Robotics came out of stealth in 2024, it seems the company makes an announcement every other week or so.

And why wouldn’t they? After all, Chef looks to be doing very well, raising a $43 million Series A, crossing 100 million meals assembled, and working with major U.S. retailers, school lunch providers, airlines, hospitals, and customers in the UK and Europe. In a space yet to see a category-defining runaway hit, Chef Robotics has slowly but surely positioned itself to possibly become just that.

All of which is why I decided to have Bhageria back on The Spoon Podcast to talk about what’s changed, what he’s learned, and where physical AI in food is heading next.

Much of Chef’s success is due in part to Bhageria making a conscious decision not to follow the same path as many other food robotics startups into the restaurant space.

Instead, the San Francisco-based startup focused on a less visible but far larger opportunity: the industrial assembly lines where millions of prepared meals are portioned and packaged every day.

Bhageria’s thinking was that while restaurant robotics that help with tasks like flipping burgers or putting cheese and toppings on a pizza are compelling, these are already tasks that can be done quickly and efficiently by a person.

“Cooking you can actually do in bulk. One person can cook for 50 people. But assembly, you actually have a lot more humans who are doing that.”

That’s because, according to Bhageria, assembly scales in a way cooking does not.

“Assembly scales linearly with output. I knew if I can automate assembly, I can actually save labor and increase volume and revenue.”

In other words, Bhageria recognized that the more meals produced, the more people are needed to portion, plate, and package them. In an era of labor shortages, that becomes a limiting factor on growth, which is why he believed food assembly was the right place to start.

That’s not to say food assembly in factories doesn’t already leverage automation. High-volume production of everything from shelf-stable packaged goods to produce relies on scaled assembly technology like conveyor belts. But one area where Bhageria believed robotics could make an immediate impact was high-variable, high-mix food production. These environments, which include everything from airline catering to grocery store prepared meals, are still largely manual. Walk through one of these facilities and you’ll see lines of workers assembling meals by hand, often in cold rooms, repeating the same motions for hours.

Traditional automation struggles in this kind of setting. Humans remain the most adaptable solution, even if they are not the most efficient.

Part of the challenge is that food is not a uniform material.

It doesn’t behave like metal parts or packaged goods. It changes constantly based on how it’s prepared, handled, and even who is doing the work.

“Onions is not onions is not onions,” Bhageria said. “If you or I chop an onion, it’s totally different. If you sauté it versus roast it, it’s totally different. And that’s just one ingredient.”

Systems built around fixed rules or specialized hardware tend to break down as soon as conditions change.

Earlier approaches often relied on dispensers, machines designed to release a specific ingredient in a specific way. But those systems struggled to generalize.

“You make the dispenser work for one kind of tomato cut… but then you cut your tomatoes differently and it gets clogged or inconsistent,” Bhageria said. “It’s very hard to scale because you need custom hardware for every little thing.”

Bhageria and the Chef team instead decided to replicate the fundamentals of how humans work, using vision, judgment, and flexible movement.

And unlike many robotics startups that created a single robot with fixed feature sets and limited variability, Chef focused early on building a system that was flexible. To do that, the company created ChefOS, a software layer that learns how to manipulate food across different environments. On top of that sits a base system built around a relatively simple hardware platform with robotic arms, cameras, scales, and containers designed to slot into existing assembly lines. Finally, the system becomes highly adaptable through its attachments, a setup not unlike a KitchenAid stand mixer, where different utensils are used depending on the task.

As the system grows and the AI-powered intelligence layer in ChefOS learns to handle ever more food types and formats, Bhageria is pushing Chef into new food assembly and service businesses. The latest of these include school lunches and airline meals, both of which operate at lower volumes than the high-mix manufacturing environments the company initially targeted.

“The next kind of step is kind of medium volume, like airline catering that might do anywhere from like 2,000 to 40,000 meals,” Bhageria said. “It’s a good step into the lower volume” business.

As Chef continues to rack up new customers and pass hundred million meal milestones, Bhageria hopes the company will eventually give the food robotics market what it hasn’t had up until this point: its own breakaway success story.

“OpenAI was that success story for LLMs,” said Bhageria. “For electric vehicles, it was Tesla. For GPUs, it was Nvidia. There isn’t that player you can kind of call out for robotics. There’s not a Facebook. There’s nothing like that for robots.”

“I always tell everybody on our team, if we can become that player, then we can inspire everybody else to do it.”

You can watch my full interview with Bhageria by clicking play below.