Close Ad

The Spoon

Daily news and analysis about the food tech revolution

Robotics, AI & Data

Shinkei Hopes Bringing Robotics & AI to the Fishing Boat Leads to Fresher Fish and Less Waste

Walk into almost any grocery store and, chances are, what you see in the fish case is not at peak freshness. 

You wouldn’t think it’d be that way, especially in places like Seattle, where I live, because such care is given to getting the fish to market quickly. But according to a new venture-backed startup named Shinkei, the critical factor in determining freshness is not what happens after fish leave the boat, but instead what happens in the moments immediately after they are caught. 

I caught up with Shinkei CEO Saif Khawaja earlier this month to discuss how exactly his company’s tech brings what he claims are Michelin-quality fish at commodity costs. According to Khawaja, conventional handling on the boat fails because it leaves most fish “flopping around,” which triggers stress responses that accelerate quality loss and shorten shelf life.

“Most fish available at a mass market retailer were handled on the boat in a way that releases stress hormone, lactic acid,” said Khawaja. “This stuff makes the meat more acidic, primes bacteria growth, and in turn speeds the shelf life and decay of meat quality.” 

Shinkei’s computer-vision-powered robot is designed to intervene immediately. Fish are placed into a machine, which the company calls Poseidon, while still alive, and it uses computer vision AI to scan the fish to determine the fastest (and least-stressful) path forward for the fish. Once the fish is scanned, the machine performs a fast sequence: a brain spike to euthanize as quickly as possible and a gill cut (to drain blood).

If this system seems, well, rough, it is, but the reality is fish caught are experiencing high stress from the time they’re caught, and the faster the fisherman can move towards euthanization, the more humane (and ultimately fresher and better tasting the fish). Khawaja says each fish is processed in about six seconds, and the company’s goal is to get the fish into the system quickly after landing, ideally within roughly a minute, before quality begins to degrade meaningfully.

Speed on the Boat = Less Waste in the Store

While much of the pitch is focused on better taste, Shinkei’s technology also a food waste angle. According to Khawaja, their solution also helps reduce waste in the store. That’s because, typically, a suffocated fish might enter rigor mortis in about 7 hours, but Shinkei’s process expands that up to 60 hours, which creates a much larger buffer before decomposition starts. Khawaja says it also makes a difference by species, with black cod handled in a traditional way lasting four to five days, where Shinkei-handled fish can stay fresh or up to two weeks. 

Khawaja attributes the compounding effect to two factors: reducing stress (less acidification) and removing blood that would otherwise diffuse through the meat and feed bacterial growth. He says the resulting shelf-life extension gives food distributors more options for logistics, allowing fish to be trucked rather than flown. 

If Shinkei’s technology works as promised, one might expect to see all professional fishermen and processors installing hardware at some point, right? Maybe…not. That’s because the company’s business model is to create a branded direct-to-consumer model for its fish, so instead of selling the hardware outright, Shinkei places machines on partner boats under a zero-cost lease and retains ownership of the machines. They also require an exclusive buying structure that grants Shinkei the right to purchase the catch processed that uses its machinery. 

From there, the company sells the fish into foodservice channels and retail under the brand Seremony, where they’re trying to get “Seremony grade” to catch on.  Khawaja says the company has sold into top-tier restaurants globally, including Michelin 3-star destinations across multiple countries, and recently launched in Wegmans (Manhattan) and FreshDirect (New York).

Today, Khawaja says Shinkei works with eight boats, sourcing species like black cod, rockfish (including vermilion rockfish), and red snapper, plus some ad hoc species (salmon, black sea bass, and others). The boats surf water in the US west coast (Alaska down to California), Texas, and Massachusetts.”

When I asked Khawaja about the underlying technology, he told me they built their AI models in-house, collecting their own data and building a pipeline informed by work like facial recognition research (fish face, that is). The computer vision stack performs a set of inferences: identifying species, detecting key points, and generating cutting paths.

He also talked about two new projects they are working on within the platform. One is Kronos, a weight-estimation model embedded in the machine that sends catch data back to the Shinkei sales team in real time so they can start selling fish before it reaches the dock. Another is Nira, which uses sensors to predict shelf life.

“We integrate sensor data into a model, and we will be able to generate ground truth at any point in the supply chain for what shelf life and quality is for that fish,” said Khawaja. 

The company recently raised $22 million and is currently at Series A. The Series A was co-led by Founders Fund and Interlagos, with new investments from Yamato Holdings, Shrug, CIV, Jaws, and Mantis.

Long-term, I wondered whether the company was open to expanding to a model in which it sells its hardware to fishermen who don’t feed their catches back to the company as part of the Seremoni pipeline.  Khawaja and Shinkei completely shut the door, but for now, they’re “focused on building the brand and basically establishing and making ceremony-grade as a certification.”

MongoDB Founder Eliot Horowitz on Building a WordPress for Robotics (And Why He’s Skeptical of Humanoids in the Kitchen)

Nearly two decades ago, Eliot Horowitz launched a company with a couple of cofounders that would be pivotal in the early days of big data, helping to create more scalable solutions, lower costs and also reducing the need for deep technical expertise needed in setting up big and messy databases. The company, which would eventually become MongoDB, also gave greater power to the developer community who leveraged the open source model to help iterate and make the product better.

Today, Horowitz hopes to apply many of the ideas that helped make him successful in big data to the world of robotics. With newest company, Viam, he is building a platform he believes will enable software engineers to more easily create and iterate software stacks for robotics and automation systems, a space where the tools to build great software are, in Horowitz’s words, “not great”.

I had heard about Viam because the company’s technology is underlies that of Gambit Robotics, a company building a vision guidance system for the kitchen. In fact, Gambit is also led by Horowitz (as well as former Google AI lead Nicole Maffeo), as a sort of a startup within a startup. In fact, Horowitz and Viam are building other vertically focused solutions on top of Viam for commercial kitchens, boat sanding, fishing and more.

I decided to catch up with Horowitz to talk to him about his move into robotics and how he sees the world of robotics development changing over the next decade. You can read an abbreviated interview below and listen to the entire conversation on The Spoon Podcast. Answers below have edited slightly for clarity.

Why did you decide to get into robotics?

And towards the end of 2020, I was frustrated with how hard it was to bring robotics projects and other hardware projects to fruition and how to make them real. And if you talk to other software engineers, their experiences were not great.

One of the topics you’ve discussed is the critical technology gap between software and the physical world of engineering. How does your new company address that?

What we really saw in this space was that a lot of interesting research had been done over the last 30 years in hardware. Batteries, motors, and robot arms have all gotten much better.

At the same time, the way people build software applications now is completely different from what it was 20 years ago. And a lot of what we saw in the hardware space is things that hadn’t been changed in a while. And it’s different, because with a robotics project, it’s both a big software project and a big hardware project. And when you have tools in the software space that don’t mesh with the hardware tools, and don’t mesh with the robotics and automation side of it, everything gets messy and just doesn’t work.

And so we said, ‘Great, what we’re going to do is we’re going to build a platform that lets engineers actually bring these projects to life.’

In the early days of the Internet, creating a website was not very easy until platforms like WordPress came along. Are you essentially making a WordPress for robotics?

It’s a really great analogy. We joke that if you dropped AWS into the hands of a Perl developer in 1998, they wouldn’t know what to do with it. We’re trying to make that same leapfrog happen in robotics. Wwe want to go from where building a robotics project is an enormous undertaking, where you have hundreds of engineers, to where you can actually build a legitimate production hardware project with real robotics and AI with an actual startup-sized budget.

What did you learn from building one of the world’s most popular database platforms in MongoDB that you are applying to your new startup?

What made Mongo great was we built an incredibly great community that we listened to almost to a fault for a long time around making sure that they were able to build exactly what they wanted and get things done

There are not enough engineers building in this space. And if we can enable more and more engineers to build in this space and really build a platform, lots of people will work together and build great things, and that’s when things get really interesting. It’s our job to enable them to do it and to show them what’s possible so that you can get a million people coming up with really cool ideas and actually bringing them to market. That’s what made Mongo great, and that’s the same thing we want to do here.

Oftentimes the time and the capital needed to bring a robot to market is just years and years and hundreds of millions of dollars. Do you believe you will shorten those timelines and reduce the capital costs by significant degree?

By orders of magnitude. What I see a lot in the robotic startups that haven’t worked is their inability to get to a somewhat interesting proof of concept (POC) that happens pretty quickly. All the engineering you have to do to make it real ends up taking years and years and years. The other big thing we see is the sheer amount of iteration time. Going from one version to the next version takes way too long.

When you can actually iterate fast, when you get a new version, and then another version, when you can test really quickly, or even iterate with users live, that’s when magic happens. And how do you bring that to the robotic space? That’s what we’re really focused on is, how do you bring that iteration time down? How do you get the proof of concept to production time dramatically down?

There are approximately 30 million software engineers worldwide, and most of them are intimidated by working with hardware. Hardware engineers typically have not had great experiences working with software engineers. It just hasn’t gone great over the last 30 years. We’re trying to completely bridge that gap so you can get hardware engineers and software engineers working together in a much more tightly knit sort of cycle and that changes everything.

What do you think about the work on humanoid robotics for the kitchen?

I’m skeptical of that. But a robot arm embedded into the back wall of my kitchen that could manage everything on the stove for me, that could stir things and add some ingredients at the right time or lower the temperature on a burner, that would be really useful. That doesn’t require any breakthroughs. We have the technology and we have the software that we could do today. So there’s a lot of middle steps (to a humanoid) that I think are very practical, very interesting, and can actually move the needle forward very quickly.

You can listen to the full interview by clicking play below or on Apple Podcasts or Spotify.

DoorDash Rolls Out Delivery Robot and an AI- Powered Delivery Orchestration Platform

There’s a new delivery robot in town, and its name is Dot.

DoorDash announced the new delivery bot today, confirming months of rumors that they were working on their own robot. The robot, developed entirely by its internal DoorDash Labs team, is the delivery company’s first homegrown bet on delivery automation and street-level autonomy, signaling the company sees owning more of the underlying robotics stack that powers delivery as a strategic priority.

Dot, which is roughly one-tenth the size of a car, can reach speeds up to 20 mph. DoorDash claims the system is designed to travel across sidewalks, bike lanes, and neighborhood roads, providing it with flexibility in navigating mixed urban environments. Watching the video of the Dot (see below), it’s clear that the robot really moves.

“(Dot) is small enough to navigate doorways and driveways, fast enough to maintain food quality, and smart enough to optimize the best routes for delivery,” said Stanley Tang, Co-Founder and Head of DoorDash Labs. “Every design decision, from its compact size to its speed to the sensor suite, came from analyzing billions of deliveries on our global platform and understanding what actually moves the needle for merchants and consumers.”

As part of the Dot rollout, the company also introduced it’s Autonomous Delivery Platform (ADP). The company describes ADP as an AI-driven dispatcher that selects between human Dashers, robots, drones, or other modes based on order type, distance, and merchant requirements.

The company described its effort to build out the autonomy stack in a post on its engineering blog. According to the company, Dot continuously ingests multi-modal sensor data (LiDAR, cameras, radar) to detect obstacles, classify terrain, and localize itself within complex urban settings. That raw sensor input feeds into a perception module, which constructs a dynamic environmental model, identifying pedestrians, street furniture, curbs, driveways, and motion patterns of nearby agents. Above the perception player lies the planning layer, which reasons about safe and efficient paths, lane transitions, sidewalk maneuvers, and mode switching (e.g. segueing from bike lane to sidewalk). Finally, the control or actuation layer translates those planned trajectories into smooth motor commands, ensuring payload stability and maintaining robust compliance with safety constraints.

In a way, the announcement of ADP as an orchestration layer is perhaps more interesting than the debut of a DoorDash native delivery robot, as it marks a big move forward for DoorDash to build a multimodal delivery network managment system. The system will streamline “handoffs today while laying the groundwork for more reliable, efficient deliveries as autonomy scales.” The company stated that it plans to work with third-party delivery technology companies (like, perhaps, drone delivery as well as other sidewalk delivery companies like Coco, with which it has already partnered).

With DoorDash’s announcement, it’s work looking at the potential impact on the delivery company’s third party technology partners such as Coco Robotics. DoorDash launched its partnership with Coco Robotics, whose bright pink sidewalk robots have been operating under the DoorDash app in Los Angeles and Chicago since 2021. While DoorDash says it plans to work with third-party providers, framing of Dot as “purpose-built for local commerce” suggests a long-term intent to shift from relying on external robot vendors to deploying its own fleet.

After the publication of this story, a DoorDash representative reached out with a comment on their relationship with Coco and other partners: “Coco is a long-term partner, offering sidewalk robot delivery for DoorDash customers in select U.S. markets – including Los Angeles, Miami, and Chicago – as well as in the EU, in Helsinki – and we’re focused on continuing to scale that partnership. Built for dense urban environments and sidewalks, Coco is designed to handle a range of delivery scenarios. With DoorDash handling millions of deliveries a day, different types of robots are required for different situations – from dense urban areas to suburban neighborhoods – which is why our multimodal strategy is so important, and why Coco is an integral part of that strategy.“

You also have to wonder if DoorDash’s move to create its own delivery bot is an answer to Uber’s strategic partnership with Serve. Some might remember that Serve, which Uber acquired when it acquired Postmates, was spun out of Uber in 2021, and that Uber still owns a stake in Serve Robotics and that the two companies work closely together.

Japan’s Most Successful Food Robot Startup is Eyeing Humanoids As The Next Big Thing

One of the defining characteristics of early successful food robots has been focus.

Whether it’s the Flippy burger-flipping robot, Bear’s front-of-house robotic waiters, or the Autocado avocado-coring and processing robot, the ability of these focused-task robots to automate one or two core functions hyper-efficiently has been a – if not the – key ingredient for success that has set these machines apart from their less successful peers.

But as the world of AI and robotics increasingly talks up humanoids as ready for primetime, are multi-function robots that more closely resemble humans – both in appearance and in their seemingly unending ability to tackle different kinds of tasks – set to take the baton as the next big thing in food robots?

If you ask Yuji Shiraki, the CEO of one of Japan’s fastest-growing tech startups and a darling of the food robotics world, the answer is a cautious yes.

The idea to build a food robot first came to Shiraki during a visit with his grandmother. Over 90 years old, Shirak’s grandmother could not cook for herself, and so he started to think about how a home cooking robot might help her. However, he soon realized that Japanese kitchens were too small to build the type of robot he envisioned, and he started thinking about building robots for restaurants. 

I first met Shiraki in 2022 at his roboticized pasta restaurant in Tokyo, E Vino Spaghetti, located across from Tokyo Station. Inside, its P-Robo robot boiled pasta, heated sauces, plated dishes, and even handled cleanup, all in just over a minute.

After the pasta restaurant, TechMagic built Oh My Dot, a ramen restaurant in Shibuya, where a robotic system prepared noodles using modular flavor packs. In both cases, the goal was to test the products with real customers while showcasing the company’s robotics to potential partners.

And the partners came. The company began working with KFC Japan, Nissin Foods (the company behind Cup Noodles), and most recently announced a partnership with Lawson.

One of the key reasons for the company’s early success was that its robots were highly tailored to specific tasks like preparing pasta, ramen, or bowl food. But now, as TechMagic and Shiraki look to the future, they see a path forward built around robots that are, like humans, much more adaptable and multi-functional.

The company outlined some of those functions in a recent announcement, saying they envision humanoids expanding human “hands” and “judgment.” Specific functions include automating repetitive tasks such as serving, sorting, and transporting food in restaurants and factories; flexible food preparation that uses AI-driven “hand technology” to perform complex cooking tasks; and customer interaction, where humanoids would optimize service and store operations using emotion and behavior recognition.

Shiraki offered clues to his bigger vision for TechMagic’s humanoid plans in a post on Facebook:

“When we began developing cooking robots in 2018, many said the ‘chances of success were slim.’ Yet today, we’ve grown to the point of competing for the top global share of operating units. The hurdles for humanoid robots may be just as high, but in the long term, we believe this is an extremely rational strategy. And beyond that, the development of humanoid robots is full of dreams and romance.”

Part of Shiraki’s motivation is to help position Japan as a leader in developing humanoid robots.

“While China and the U.S. are leading the way, we intend to contribute to labor-strapped industries with a Japan-born humanoid robot and expand globally.”

He says the company is hiring and looking for partners. One of the first steps is working with existing humanoid robots such as those from Unitree in the TechMagic development lab. Shiraki even showed off a video of the Unitree robot on Facebook.

Shiraki told me they are eyeing around a three-year time horizon to develop their first humanoid for the food business and that the company is now busy raising its next funding round to help fund the development.

“TechMagic is taking the technology it has cultivated in cooking and service robots to the next level, fusing it with humanoid robotics to create a new ‘future of food’,” said Shiraki. “We aim to build a social infrastructure that frees people from boring, harsh, and dangerous tasks, enabling them to live more creatively.”

Drone Delivery is About to Get a Big Upgrade. Here’s Why Part 108 Will Change Food Delivery Forever

For those of you who are skeptical about whether drone delivery will ever become a common way to deliver your pizza or groceries, I’m there with you. Several key factors need to be in place to ensure drones can deliver items quickly, at low cost, and, perhaps most of all, safely.

One of the key hurdles to ensuring all of those things become a reality happened last week with the release of Draft Part 108 for Beyond Visual Line of Sight (BVLOS) by the FAA. Part 107, introduced in 2016, set the baseline rules for small commercial drone operations, allowing flights within visual line of sight under a certified remote pilot but leaving BVLOS missions dependent on case-by-case FAA waivers.

The new proposed rule brings with it a framework for BVLOS operations that replaces ad hoc approvals with standardized pathways. For industries like food delivery, the new rules could be the regulatory green light that turns pilot projects into citywide services.

Individuals to Organizations: One of the biggest changes from Part 107, which put the burden on individual pilots (requiring each to be certified and limiting most flights to visual line of sight), is that Part 108 shifts accountability to the operating company rather than every drone pilot. This change acknowledges that modern drone delivery (especially BVLOS), relies on automated systems, centralized oversight, and coordinated fleets, not one pilot per aircraft.

Permits vs Certificates: Operators will choose between two regulatory paths designed to match the scale and complexity of their operations. Permits will work for lower-risk, smaller-scale flights in less densely populated areas and come with a cap on the number of drones (such as a limit of 100 for delivery services). Certificates are geared for high-density, high-volume operations in urban environments, and will remove fleet caps in exchange for stricter oversight, safety requirements, and operational protocols.

Operations will be governed by five population density tiers, from rural (Category 1) to major metro cores (Category 5). Permits only allow access to Categories 1–3; Certificates are required for Categories 4–5. For food delivery, that means suburban and exurban rollouts first, with dense city markets requiring the more stringent certificate process.

A New Player: ADSPs. One of Part 108’s biggest changes is the creation of Automated Data Service Providers which are companies responsible for real-time airspace data. This data includes:

  • Drone traffic and location tracking.
  • Weather integration.
  • “Strategic deconfliction” to prevent midair conflicts.
  • “Conformance monitoring” to ensure drones follow approved routes.

Operators must connect to an ADSP, but they can also become their own ADSP if they meet the technical requirements.

There are a bunch of other changes – I recommend you check out this great writeup by Matt Sloane. For drone food delivrey, it will bring big changes, helping to take drone delvery from what is essentially a novelty service today to a scalable service.

The primary bottleneck has been regulatory: BVLOS flights, essential for covering enough ground to make delivery viable, were locked behind an unpredictable and time-consuming waiver process under Part 107. Part 108 changes all that. By creating clear, scalable pathways (permits for smaller suburban rollouts, certificates for full-scale urban operations), the FAA is giving food delivery companies a roadmap to expand without reapplying for exemptions every time they add new routes or drones. The introduction of Automated Data Service Providers means operators will have access to the real-time airspace management needed to safely run dozens or even hundreds of flights at once.

Add in the push toward autonomy, standardization of safety measures, and population tiering, and Part 108 looks less like the regulatory green light that could take drone food delivery from novelty to serious delivery option.

As Robot Delivery Grows, Buildings May Have “USB Ports” to Enable Seamless Delivery Handoff

With this week’s news that Little Caesars pizza is partnering with Serve to deliver pizza, it’s clear that robotic delivery, whether on sidewalks, underground, or in the air, is fast becoming more mainstream.

Despite many of these startups working on technology for close to a decade, the robot food delivery process still has a cumbersome part: picking up the food at the restaurant or handing over to the customer in their home or apartment.

The main reason is that most restaurants and homes were not built for robotic delivery. Sidewalk delivery robots and drones can’t open doors or use elevators, requiring customers to visit designated pickup spots or follow instructions to collect food at the curb.

This might change in the future, according to Garrett McCurrach, CEO of Pipedream Labs. McCurrach believes that in the future, buildings will adapt to robotic delivery. “A lot of buildings in the future are going to have what is essentially like a USB port for accepting and handing off to different modalities,” said McCurrach at SKS 2025.

According to McCurrach, the goal on the restaurant and ghost kitchen side is to solve the coordination challenges of last-yard logistics, especially in environments where speed and efficiency matter most. “Being able to get things out quickly into the right modality, whatever the modality is, is going to be super important,” he said.

Will Buildings Have 'USB Ports' for Robot Delivery in the Future? This Founder Thinks So.

While drone landing pads and delivery hubs have already been deployed in the early stages of the drone delivery market, some are exploring how drones could deliver vertically within a building itself. A group of researchers at the University of Tokyo looked at potential scenarios of drone delivery in high-rises, essentially through what is a “USB port” at the top of the building.

Looking further into the delivery future, another potential big challenge is enabling multi-model (or multi-robot) delivery. Sure, delivering via a drone might make sense to some places, but a more realistic option might be drone to sidewalk robot or, eventually, a humanoid.

For McCurrach and Pipedream, he’s had to think about handoffs to other automated forms of delivery since the beginning, in large part because his delivery system – which consists of an underground delivery with specialized delivery robots – almost by definition will not reach the end-users home or work since the company’s network only delivers to drop-off points in building lobbies or other specialized locations.

“We build the underground networks to move things quickly to get to that end node. And then portals that hand off to a self-driven car, a Coco robot, a Zipline drone, whatever the right modality is to get it to its end destination,” said McCurrach.

But, if early USB technology is any example, over time, universal connections via coordination across ecosystems become more prevalent, as both physical infrastructure and robot-to-robot coordination increase, I expect the arrival of multi-modal robot delivery to be set to arrive at our doorstep sometime in the next decade.


Is Posha the Robotic Heir to the Thermomix? The Founders Sure Hope So

For the past five or so years, the emails have landed in my inbox on a steady basis, nearly every month. They’ve included updates on a small startup building a countertop cooking robot named Posha.

The emails, almost always written by the company’s CEO and co-founder, Raghav Gupta, detail progress, both big and small, ranging from software tweaks and field trial insights to news of an $8 million Series A funding round.

The progress has been slow but steady. And over the past year, the company has reached a milestone that most cooking robot startups (especially those targeting the home) haven’t come close to: they’re now building robots using scaled manufacturing, and those robots are arriving in customer homes.

Given that I’ve followed dozens of companies attempting this goal over the past decade, I figured I’d take Raghav up on his invite to see the robot in action and talk with him about what’s next.

So this past Sunday, I headed to an Airbnb Raghav had rented north of Seattle to demo the Posha robot for media and investors. Raghav asked if I wanted to cook a meal with Posha, and within minutes of arriving, the robot was preparing spaghetti Alfredo.

The machine stirred, heated, and timed each step with minimal interaction from me. Posha includes four ingredient containers, multiple spatulas, a spice carousel, and an induction cooktop. A camera watches over the food, analyzing “color, texture, consistency,” and, according to Raghav, provides “human chef-like intelligence.” Users load chopped ingredients, select a recipe, and let the device do the rest. “You just tell Posha you need that, and you walk away,” Raghav explained.

Posha, originally named Nymble (both the robot and the company), has changed significantly from its early days as a college project. “We were two people taking out of our parents’ garage trying to make a cooking robot.”

The first version was a robotic arm, but Raghav said customer feedback led them to pivot. “We had this choice of either repurposing our robotic arm for commercial kitchen use cases or changing our technology altogether to make something that consumers wanted. We chose the latter route because we were in love with the problem we were trying to solve.”

That problem: helping people figure out what to eat on a daily basis. “People like you and me want to eat freshly cooked meals and feed our families freshly cooked meals. But it’s hard to find the time to cook these meals every single day.” He believes this tradeoff, between eating well and having enough time, is what led to a national health crisis. “We are in the middle of a health catastrophe,” he said. “And I think with Posha, it will help America become one of the healthiest countries in the world, at the same time being one of the most productive countries in the world.”

Those are lofty goals, ones I’m pretty skeptical about given the high price tag of the Posha and the nearly non-existent adoption of cooking robots so far. But according to Raghav, he sees his product as a natural evolution of a device that has been quite successful, especially in Europe: the Thermomix.

“I think we have a strong precedent in terms of Thermomix. They sell like a million units every single year, and what Posha is, is actually Thermomix++.”

If there’s a model to aim for, the Thermomix is a good one, and I have to say, the ease with which I was able to make spaghetti Alfredo was reminiscent of the first time I used a Thermomix. In fact, it was essentially what Raghav described, the Thermomix++, in that it required me to do even less once I picked the recipe and hit go. From there, over the next 30 minutes, the Posha added ingredients and cooked the meal to completion.

It’s perhaps this ease of use and the similarity to Raghav’s professed North Star in the Thermomix that helped the company recently raise over $8 million in Series A funding. You’d have to be living under a rock, covered with more rocks, and then some dirt not to realize how hard it is for consumer hardware startups to raise money (let alone a robot cooking startup). The fact that Posha secured funding led by Accel is a sign they may be doing something others in this space haven’t.

So far, Posha has shipped 200 units, with 600 more expected by the end of September. “We’re trying to grow 3X every six months or so,” Raghav said. The product retails for $1,750, with pre-orders at $1,500.

If you’d like to see Posha in action, check out my cooking video below. Raghav will also be speaking about his journey at the Smart Kitchen Summit this week, so if you want to hear more and ask him questions, make sure to grab your ticket..

The Spoon Cooks a Meal with Posha the Home Cooking Robot

From Aspiring Pro Surfer to Delivery Robot CEO with Coco’s Zach Rash

Zach Rash wanted to be a professional surfer. So much so, that in high school, there was more surfing than academics.

That all changed when Rash reached UCLA and met Brad Squicciarini. It wasn’t long before the two spent every waking hour together in a small room building robots.

“We spent like our entire life in this like box at UCLA with no windows, and we’re just building robots from scratch, and it was the best job ever.”

Eventually, the real world came knocking as Rash and Squicciarini graduated and had to find jobs. After applying for many of the same positions, they eventually decided they should just start their own robot company.

“We just had a lot of really strong opinions about what it would take to get these things into the world and make them useful. So… decided to do it ourselves.”

Coco launched in early 2020. “We started building them in our living room and we couldn’t get more wheels… so it was a bit of a sketchy robot.” Still, their first merchant deployment went smoothly. “The first day of the business, I mean, we gave it to a merchant and Brad and I just took turns driving it and fixing it.”

They faced steep financing challenges: “We didn’t have any money… Even if you’re only building a few, you know, it’s still going to cost you tens of thousands of dollars.” They pitched more than 200 investors before raising a modest $50,000 to start. “We thought that was a lot of money and we built a few robots with that and kind of proved out that we could run a service, not just build the robot.”

Their persistence paid off. In June 2025, Coco raised $80 million, led by angel investors Sam Altman and his brother Max, alongside Pelion Venture Partners, Offline Ventures, and others. 

This brought Coco’s total funding to over $110 million, which Rash says the company plans to use to scale its operations and technology.

“Coco Robotics will use the new funds to improve the technology and to scale up its fleet,” Rash told TechCrunch. “The company expects to go from low‑thousands to 10,000 robots by the end of next year.”

According to the company, Coco bots have delivered over 500,000 items to date, working with retailers like Subway, Wingstop, Jack in the Box, Uber Eats and DoorDash.

It’s only been a few short years since Rash was largely concerned about surfing, but now, armed with funding and lots of interest from retail partners, he’s ready to ride to the wave of growth of his robot delivery company.

“We’re building as many as we can as fast as we can.”

Zach will be speaking at SKS 2025 tomorrow, so make sure to get your tickets. You can listen to our conversation on the latest episode of The Spoon podcast below, on Apple Podcasts or Spotify, or wherever you get your podcasts.

This Culinary Tech Inventor Thought He Could Build Some Parts For His Latest Gadget in the US. Then He Called Around.

When kitchen tech inventor Scott Heimendinger started prototyping his latest hardware product, he knew that much of it would need to be built overseas. Still, he was curious whether he could rely on local Seattle-based shops to produce some of the parts.

“I contacted local shops here in Seattle. There are a couple of machine and metalworking shops, and I thought, well, I would like to be a good customer, right? Like I’d love to spend money locally, especially on shops that are doing this kind of stuff.”

But when he called around, Scott quickly discovered that not only were the local shops going to be an order of magnitude more expensive, but they’d also take longer to deliver.

“I said, ‘look, I know this is going to be more expensive than what I’m doing in China, obviously, but maybe we can make this up on the time front.’ Before we even got into real pricing, we were already above 10X. So I said, ‘What about turnaround time?’ [They] said, well, it depends how busy we are, but like, you know, one to six weeks.’”

We’d started talking about the cost and complexity of building in the U.S. because we’d both recently listened to an episode of PJ Vogt’s Search Engine, in which Vogt interviewed YouTuber and engineer Destin Sandlin. Sandlin discussed his years-long effort to manufacture a product in America, and I wanted to get Scott’s take, especially since he’s been navigating the uncertainty caused by new tariffs. As it turned out, he had a lot to say.

One area he pointed to as a critical missing link was the shortage of tooling designers, the specialists who create the molds used to shape plastic parts.

“Tooling fabrication in principle is something that you could just do on a beefy CNC machine… In practice, no. It’s specialized techniques and tools. That knowledge has dried up in the U.S.”

We talked about why capabilities like tooling fabrication and injection molding have largely disappeared from the U.S., and one reason we both agreed on was the lack of trade education, starting as early as high school.

“Some of my favorite classes in high school were sculpture class, learning to use a bandsaw and a drill press,” he said. “I wish more folks in the United States prioritized the hands-on making of stuff.”

I pointed out the strange dichotomy of the past couple of decades, in which Silicon Valley was busy valorizing the maker movement, while at the same time the U.S.’s ability to manufacture at scale was simultaneously being hollowed out. It’s as if we celebrated prototyping, while the infrastructure to mass-produce those ideas was quietly de-emphasized and disinvested in.

“A weird thing that happened, where we talked about, ‘hey, let’s start making stuff and teach our kids to make stuff,'” I said. “But at the same time, America’s ability to make stuff at scale just kind of went up in smoke.”

Scott, for his part, chose to see the upside. Despite the loss of critical manufacturing knowledge and infrastructure, he said it’s still a great time to be an inventor, thanks to how accessible prototyping tools have become.

“I’m of two minds about this. On the one hand, like I love physically making stuff. I wish more folks in the United States prioritized the hands-on making of stuff, and I wish that we hadn’t eroded away these capabilities. On the other hand, it is almost point and click to have these things prototyped, if not mass-produced. And that’s an incredible boon to being a scrappy solopreneur.”.

You can listen to our latest episode by clicking play below, or you can find it on Apple Podcasts, Spotify, or wherever you get your podcasts.

If you’d like to ask Scott a question about his project, the challenges of manufacturing a product or the future of cooking, he’ll be at Smart Kitchen Summit next week. You can get your ticket here.

If video is your preferred podcast consumption format, you can also watch our conversation below:

Why It's So Hard to Build Hardware in America

Thermomix Has Long Been a Leader in Cooking Automation, But Now They’re Going Full Robot

For years, I’ve said that the Thermomix is quite possibly the most successful automated cooking appliance in the world. Sure, it’s not a humanoid robot or what we’ve come to expect from cooking robots in recent years, but the TM6 and TM7 are software-powered cooking appliances that automate and sequence functions in a way that feels surprisingly intelligent, especially compared to typical countertop or built-in kitchen appliances.

But now, if recent moves by Thermomix’s corporate parent, Vorwerk, are any indication, Thermomix may be going full robot. At last month’s Automatica conference in Munich, Thermomix and red-hot German robotics startup Neura Robotics announced a partnership in which Neura’s humanoid robot used Vorwerk’s Thermomix and Kobold vacuum cleaners to perform everyday household tasks.

According to Neura CEO David Reger, optimizing his robots to work with Vorwerk’s cooking and cleaning appliances is a step toward building an aging-in-place platform powered by humanoids.

“Together with Vorwerk, we are redefining household robotics – with intelligent assistants that provide concrete relief for people in their everyday lives: from cooking to independent living in old age,” said Reger.

Even more interestingly, Vorwerk also announced a partnership with AI and chip giant NVIDIA last month. According to the announcement, “Vorwerk is post-training NVIDIA Isaac GR00T N1, an open robot foundation model, to support families around the home, whether seniors looking to maintain their independence, or busy families in need of an extra pair of hands. To post-train the model, Vorwerk is leveraging the Isaac GR00T-Mimic data pipeline to generate large, diverse synthetic motions data to prepare robots for common household tasks such as cooking, cleaning, and more.

“Together with NVIDIA Robotics we are now taking a significant step towards the connected and automated home,” wrote Vorwerk CEO Thomas Rodemann on Linkedin. “Our goal: creating integrated digital/physical ecosystems that support our community in their everyday lives and make the home more convenient for everyone – whether it’s providing busy families with an extra pair of hands or giving seniors more independence.”

When Jensen Huang showed up at CES in January and said that the ChatGPT moment for robotics is right around the corner, I’m not sure if he was thinking about cooking food with the Thermomix, but maybe he was. Vorwerk would be a logical candidate to build true home robot assistants, since progressing rightward on the simple tool to fully-capable robot continuum already and they’ve been the most successful at integrating software with home cooking automation

You can watch the video of the NVIDIA-powered robot making food with the Thermomix in the video below.


Food Assembly Robot Startup Chef Robotics Raises $43M Series A as it Reaches 40 Million Meal Milestone

Chef Robotics has raised $43.1 million in a Series A round to accelerate deployment of its AI-powered food assembly robots, the company announced today. The funding brings the San Francisco-based startup’s total capital to over $65 million, including equipment financing. Avataar led the round, with participation from Construct Capital, Bloomberg Beta, and others.

Founded in 2019, Chef Robotics is building what founder and CEO Rajat Bhageria calls an “AI platform for food.” Rather than building a single-purpose robot, Chef’s system is designed to work in diverse food production environments—learning and adapting through software to new tasks like portioning, topping, or filling.

When I first got a peek at Chef’s system last year, I was intrigued because the company had struck a balance that seemed to elude many food robotics startups. While startups in this space seemed to make either high-volume solutions with limited customizability or use off-the-shelf robotic arms that aren’t made for true high-production, Chef has built a flexible and scalable robotics platform that can be customized for any number of high-volume food production environments.

That’s because while many robotics companies focus primarily on hardware, Chef’s approach centers on a software layer that enables “Embodied AI”—giving physical robots the intelligence to operate autonomously in real-world conditions. Chef’s system combines a robotic arm with AI models trained on millions of real-world examples. These models, powered by production data from early customers like Amy’s Kitchen and Fresh Prep, allow the robots to generalize across new ingredients and dishes. To date, Chef Robotics has helped assemble over 40 million meals.

From the company’s announcement: When we thought about starting with restaurants, we ran into the chicken and egg problem – to enable robots that are flexible enough to add value, we need a highly capable AI, but to get a highly capable AI, we need real-world training data from the customer sites…. Thus, we decided to initially deploy robots in high-mix (read as highly flexible) food production and manufacturing environments where Chef could partially automate a food operation and thus add value in production to customers without requiring 100% full autonomy from the get-go. We built Chef’s systems on modern advancements in AI to make them highly flexible and adaptable enough to “pick” and plate almost any ingredient, no matter how it’s cut, cooked, or grown; this makes them an ideal solution for assembling or plating food.  

The new capital will support scaling up deployments and building out Chef’s sales and marketing teams. The company is currently active in the U.S. and Canada, with plans to expand into the UK next year.

Bridge Appliances Deploys Egg-Making Robot at First Customer

Five years ago, the cofounders behind Bridge Appliances stood in line at a busy breakfast cafe. As minutes ticked by, frustration turned into inspiration. They wondered: What if the preparation of eggs could be automated? That simple question led to the creation of OMM, a countertop egg-making robot. Now, half a decade later, the Bridge team is back in a coffee shop, deploying their robot for their very first commercial customer, Beantrust Coffeebar in Beverly, Massachusetts.

Bridge cofounder Connor White recently described spending the past two months embedded at Beantrust, collaborating closely with owner Erik Modahl and his team. According to White, working alongside baristas, listening to customers, and absorbing the café’s unique culture and operational flow allowed Bridge to tailor OMM’s integration precisely to Beantrust’s specific needs.

As I wrote last year, OMM cooks two eggs in roughly two minutes, enabling Beantrust to serve around 60 eggs an hour. White notes they’ve already seen promising results, with the new sandwich lineup boosting average ticket values by 15%. Currently, one in five customers chooses to add a freshly made sandwich to their coffee order (and that number continues to climb).

This marks a significant milestone for Bridge, which raised $2 million in seed funding from Steve Papa, one of Toast’s earliest investors, in 2021. Moving forward, Bridge is likely to see more growth among small coffee shops or similar establishments that lack full kitchens or grill cooks but still wish to offer breakfast. However, they will need to raise considerably more funding to scale effectively, or they could be a potential attractive acquisition candidate for a company such as Middleby.