If food scaled like software, we’d all be unicorns by Q2. But food has factories, freight, and fickle taste buds. It’s made of atoms, not bits.
Adam Yee, food scientist and founder of Umai Works, makes flavor formulas and food prototypes for the next wave of brands. He’s measured enough to know where AI can help, and why food can’t (and shouldn’t) try to run like software.
For people who don’t know you yet, can you give a quick introduction?
Hey, I’m Adam Yee, I’m a food scientist, and serial entrepreneur. Previous companies include Better Meat Co and Sobo Foods, which are both venture-backed companies. Along the way, I’ve worked in a few corporate companies ranging from protein bars, supplements, to precision fermentation. I work on a lot of cool projects under Umai Works, from AI projects to solving hard technical challenges to helping bring ideas to life.
I also do podcasts such as My Food Job Rocks, a weekly podcast where we interview professionals in food, Maybe Food Maybe Tech, a weekly news commentary podcast on technology (like AI) and food, and Food Products FAQ, a 10-part mini series about what we as food scientists look for when developing food products..
You’ve experienced the challenges of scaling first-hand. How is AI making that process faster, cheaper, better?
It’s important to point out that we really don’t have a clear value proposition for using AI in the food industry. The only people who seem to say AI helps with something in food also want to sell you a $5,000 program.
In fact, at this stage, we see AI as more of a marketing tool to raise money than actual effectiveness.
An MIT sloan report states that despite $30–40 billion investment into GenAI, “a surprising result in that 95% of organizations are getting zero return. The outcomes are so starkly divided across both buyers (enterprises, mid-market, SMBs) and builders (startups, vendors, consultancies) that we call it the GenAI Divide. Just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact. This divide does not seem to be driven by model quality or regulation, but seems to be determined by approach.”
This basically means that unless you are a very certain, very progressive, and very flexible organization, the chances of an AI program actually working is low. The food industry is historically NOT any of those things.
However, I think there’s a lot of promise if companies can clearly show the value of AI. One company I interviewed on My Food Job Rocks was Starday, which uses an AI brain to analyze market trends to create retail products, which I thought was a really interesting and practical take.
Other companies like Alphafold, which can potentially use AI to crunch numbers and show us new innovative forms of proteins, is also something I’m excited about.
How close are we to developing new recipes without ever stepping foot in a kitchen?
According to my friends in Supply Side West, there are more and more companies trying to do this. However, most of these things right now can be done with ChatGPT and the right prompting.
I was fortunate to be recruited by Food Systems Innovations and Stanford University to evaluate how LLMs like ChatGPT could help with food scientists. We ended up taking the nutritionals, the ingredients, and the sensory results of about 20 plant-based meat products and trialed a whole bunch of different prompts.
We found that ChatGPT will give a good experimental design you could use to improve your product based on sensory feedback, which I think is really cool. However, more research is needed to see if the experiments will actually work. [Link to the published paper.]
We received a $2 million dollar grant from the Bezos Foundation to continue this work so I’m really excited to get more exposure on how we can improve our understanding of using LLMs in food.
I think we are scratching the surface on how this technology will affect food companies. I also believe there will be a lot of bad tools made by non-food industry firms out in the market that will put a stain on using this technology. It always happens that way.
What’s the biggest disconnect between how founders imagine food innovation and how it really plays out?
I think there’s an overall huge disconnect in how people think food is made versus how it actually is. Our perception of scaling that comes from software has really warped our hopes and dreams that a food company will scale like a tech company. This was probably the biggest flaw in what we dub as “food tech” (basically a food company that can raise at tech valuations).
I lived through this working at Motif Foodworks, which raised $345 million dollars and suffered an ill fate. However, it was more-so a biotech company in which money like that is lost all the time. But the same thing happens: money is raised, hubris kicks in, and things get wild, then get very painful.
One of the core lessons I learned is complexity tends to bite you in the ass. Anything complex about a food product will exponentiate in cost and time as you scale. It’s important to be flexible when it comes to simplifying your product, but it’s a hard decision whether to keep or remove those complexities.
Food has always struggled with the traditional and the technological. I always recommend to think simple first, and then maybe evolve as you build it. But if complexity is your secret sauce, you better figure out every way for it not to screw you in the future.
Which emerging tech is ready for CPG to use today?
You’d be surprised how low-tech most CPG companies are. It’s a low-margin business that makes technology less of an importance than you think. A lot of companies I work with don’t even use Excel correctly. There’s just too many things that a food entrepreneur has to worry about besides technology.
I would only consider software or tools when you start to hire employees. I don’t like recommending tools because each fits a specific use-case. Currently, the only AI tool I would recommend is ChatGPT (the $20 dollar version).
Any food+bev CPG companies you see using tech the right way as part of their process?
In the CPG industry, not really. Everyone big is worried about tariffs and GLP-1, and everyone small is worried about getting bullied by their retail distributors.
However, I think those that focus on more boring parts of the food industry are using AI a bit more effectively. For example, food safety and regulatory, which has a lot of firm rules and very concise data, are testing using LLMs in really creative ways, from devising concise SOPs to crunching the numbers on microbe counts.
If you were launching a food brand today, how would you bake in technology from day one?
In my opinion, the only way someone is going to succeed with technology, and specifically AI, is that the technology has to produce a noticeable and successful output, which causes other people to hack and solve how they did it. Eventually, they will find an AI program with the same capability.
We see this playbook all the time in the ingredients industry: there is a viral product (like allulose or epogee), food scientists do their reverse engineering bit, and learn the product is successful because a certain ingredient is used in a unique way.
I think the same will be for AI tools, but it will be difficult to find a direct correlation between a startup’s success and a certain AI tool. I don’t think it will be product-specific, I think it will be based on the outputs of successful products at a speed we would find impossible.
So launching a food brand today, I would focus on how you get as fast an output as possible using technology and test, test, test.
Adam Yee is a food scientist, serial entrepreneur, and podcaster. From 4 a.m. shifts in a granola bar factory to founding venture-backed startups like Better Meat Co. and Sobo Foods, he now leads Umai Works and helps turn big food ideas into real products.