August 28, 2026 | The Product Manager’s Journal
The Take
This week, a food-tech company raised funds for software that can create a product formula before anyone runs a physical test. Work that once took months in the lab can now take days. At first, that sounded useful. My second thought is the one I want to share.
If a model can suggest the recipe, predict how it will turn out, and reduce the number of tests needed, what is left for the product manager to do?
It turns out there’s more left than I thought. Making experiments cheaper doesn’t make good judgment less important. It just makes it easier to repeat bad decisions. When a new formula costs ingredients, lab time, and a week of someone’s focus, that price forces teams to narrow their choices first. Remove that cost, and suddenly you can create twenty answers before anyone agrees on what makes a good one.
That’s the risk. The model gives you results that look certain, and the team might treat a winning formula as a final decision. But every recommendation comes with a tradeoff. Look closer, and you’ll see the cost. Maybe the cheaper blend changes the texture. Maybe it means losing a regional supplier. Maybe it makes production easier but the story harder to explain.
So now, before I accept an AI-generated blend, I ask four questions. Which constraint is real, and which one is just habit? What did the model optimize, and what got worse because of it? Who owns the tradeoff, since a model can suggest but can’t be accountable? And does the answer still taste like us? That last question won’t appear in an optimization score, but it’s the most important.
Here’s what I’m reminding myself: taste is saying yes to just a few things. AI makes it easy to say yes to everything. The real job is protecting the no.
Spice Route Signal
Big food companies are investing heavily in AI to personalize food for your biology. That market is real. But it’s a race I’m choosing not to join, and knowing which race you’re in is half the battle.
The AI-driven personalized nutrition market is expected to reach about $5.55 billion in 2026, up 23 percent from $4.5 billion in 2025, with forecasts projecting it to reach $12.75 billion by 2030. Danone’s OneBiome Laboratory in Paris-Saclay is using microbiome data, clinical research, and AI to study how gut microbes affect health and how food can support it. On the startup side, direct-to-consumer kits analyze your gut flora and send you a personalized diet plan.
Notice what all of this is focused on: a health score or a biomarker result. That’s a different product from what Trevean offers, and the distinction matters more as the biomarker space gets more crowded and better funded.
Trevean focuses on flavor and cultural discovery. Our recommendations are a way to explore a cuisine, not another way to manage your body. While other companies compete to interpret your biomarkers, we can help you taste somewhere new. That doesn’t sound small to me. It sounds like open space.
The product manager move isn’t flashy, but it works. Write down which race you’re not running. Then review your packaging, pricing, and pitch for any words that quietly pull you into it. Borrowed positioning falls apart quickly. Your own has to hold up.
[Sources: AI in Personalized Nutrition Market Report 2026, The Business Research Company; Danone OneBiome Laboratory and Danone’s microbiome research overview.]
From the Trenches
Last week, I had just one finished lid. This week, the production run began. It’s the first time all quarter that a physical part has passed approval at my desk.
The lid maker sent photos and videos of the sample run, then started production. After a year of drawings and measurements, we finally have a real part, and more are on the way. It might sound minor, but it didn’t feel that way.
Progress happened on the tap side too, but this time it came through decisions rather than samples. I had my first recorded meeting with our NFC developer, and in twenty-two minutes we made more progress than in some entire weeks. We’re moving DNS management to a dedicated service while keeping the domain registration where it is. The blend pages will live on our own host, and the NFC tags will point to those pages rather than to a site we don’t control. That fixes a problem I found two weeks ago. One blend’s content is almost ready; the other is waiting on final photography before it can ship.
On the sourcing front, the second co-packer is now a real option instead of just a possibility. A few weeks ago, if one supplier stopped responding, everything would have stalled. This week, the new co-packer walked me through blending, packing, safety seals, and label application. All he needed next was photos of the labels to confirm they would work on his machines. Redundancy isn’t glamorous, but it turns a single point of failure into a choice.
Here’s the honest part. I did my own mid-quarter review, and it was uncomfortable but useful. More than half the quarter is gone, and almost all of it went into physical and technical parts: lids, chips, hosting, co-packers. The investor story and the one standout asset that would make the product clear in a room still aren’t there. Not even a draft or a storyboard. I’ve been shipping the parts I know how to handle and avoiding the one I don’t.
There’s still a small invoice unpaid and a domain step half-done, both fifteen-minute tasks that keep getting pushed to next week. It’s the same lesson as always: the boring task you avoid is the one that will block the demo when you need it most.
So this week brought me a real part, a clearer tap setup, a second supplier, and a lesson I didn’t want. Progress and a warning, all in five days.
The Spice Lesson
Rose gives Persian Sunrise one of its most distinctive notes, but add a little too much, and the whole blend starts tasting like perfume.
Rose is one of the quieter ingredients in Persian Sunrise. It adds a soft floral note that lifts the blend and makes it unmistakably Persian. But rose has a narrow working range. Use too little, and it disappears. Use too much, and it takes over, turning food from fragrant to soapy. The difference between distinctive and distracting is smaller than you think.
That’s the product lesson hiding in the petals. We often think more is safer: more features, more requirements, more of whatever tested well. But the ingredient that gives a product its identity can also overwhelm it. The team that adds one more requirement, just to be safe, is often the one that buries what people liked in the first place.
The hard part is finding the point where the idea is clear and then stopping. Decide what your product is really for, add enough for people to notice, and resist the urge to make it louder. Restraint isn’t the absence of a feature. Sometimes restraint is the feature.
On My Desk
“Essentialism” by Greg McKeown.
McKeown’s main idea is simple but powerful: the disciplined pursuit of less. It isn’t about doing less for its own sake. It’s about focusing your effort on the few things that matter and saying no, clearly and often, to everything else.
I keep returning to this idea during busy build weeks like this one, when it’s easy to do a bit of everything and finish nothing. A tool that gives you twenty options is an essentialism problem waiting to happen. The tool makes it easy to say yes, so the real job is protecting your no. If you don’t set your own priorities, someone or something else will. This week, that “something else” was a long build list.
That’s the Rack
Thanks for reading Issue #22. I’m Dan Blizinski, founder of Trevean Spice and the writer behind The Product Manager’s Journal. I focus on PM frameworks that come from real experience building things, not just theory.
If you’re new here, check out the free Startup PM Toolkit. It includes five frameworks I actually use, not just talk about.
What’s the one thing you keep adding just to be safe—the thing that might actually be making things worse? Reply and let me know. I read every message.
The PM’s Spice Rack is published weekly on The Product Manager’s Journal and on LinkedIn. Subscribe to get it in your inbox.
Leave a Reply