The Nutrient
Nutrition, AI, and building in the open
Short, honest notes on where nutrition tech is heading — and what we're learning while we build Mr BITE. New issues land here first.
-
The Nutrient Our Recipe Database Doesn't Know What 'Ultra-Processed' Means Either HHS and USDA submitted the federal government's first proposed definition of ultra-processed food on August 10th, and industry groups are already fighting to slow it down. Mr BITE's own barcode lookup queries the exact database that already scores processing level, and our code explicitly leaves that field out of the request. -
The Nutrient Our 'Keto' Preset Isn't the Diet the New Trial Tested A trial published August 27th in Cell Metabolism found a strict ketogenic diet, just 4% of calories from carbs, cut liver fat 67% and reversed prediabetes in half its participants. Mr BITE's own Keto preset lands closer to triple that carb share before you've even counted the grams. -
The Nutrient Our Protein Targets Never Ask How Old You Are A widely debated new review argues protein needs should be personalized by age and life stage, not one flat number for every adult. Mr BITE already collects a user's age to calculate calories, then drops it completely before deciding how much protein they should eat. -
The Nutrient The Feature That Would Prove You're Succeeding Is Locked Behind the Wrong Metric A new JMIR study of 5,132 food-tracking app users finds that logging 2-3 days a week, not perfectly, still predicts real weight loss. Mr BITE's own Progress tab stays locked until a user strings together 7 straight days, the one pattern that study says isn't what matters. -
The Nutrient The Keto Preset Doesn't Know What Your Genes Already Decided New Stanford research presented at NUTRITION 2026 finds a low-carb, high-fat diet spikes LDL cholesterol sharply in people who are genetically predisposed to it, and barely moves it in others. Mr BITE's Keto macro preset, 80 percent of calories from fat, is one tap away from any user, with no cholesterol or genetic screening standing between them and it. -
The Nutrient The Protein Number Was Never the Hard Part A new Yale-led study finds GLP-1 drugs may reprogram metabolism rather than just suppress appetite, and separate research shows most people on them fall short on protein anyway. Mr BITE's own protein formula already lands in the right range. Our safety logic guarantees the people who disclose a GLP-1 never see it. -
The Nutrient The Confidence Label Was Never a Substitute for a Scale A NUTRITION 2026 study found four popular AI photo-calorie apps underestimate meals by roughly a third against real, weighed ground truth. Mr BITE already tags its own photo estimates with a confidence label, and we've never tested whether that label means anything. -
The Nutrient The Calorie Total Was Never the Actionable Number A JAMA Network Open trial found a low-fat vegan diet cut people's dietary energy density by nearly a third, and they lost weight without eating less food or counting a single calorie. Mr BITE logs the exact numbers that measure would need and has never shown it to anyone. -
The Nutrient Grounding an AI Coach in Real Data Is the Easy Half Emory's new NutriCamp app estimates 65 nutrients by retrieving from USDA's own database instead of trusting a model's memory, which is exactly the right call. Our week proved the harder problem: telling a grounded answer from an invented one after the fact. -
The Nutrient Every App Hands Out the Same Protein Target, and the Science Disagrees A 350-paper review out of the University of Wisconsin-Madison found protein needs hinge on how active someone actually is, not age or weight alone, which most nutrition apps, Mr BITE included, don't yet act on. -
The Nutrient The Rules of Thumb Baked Into Every Food Score Just Had a Bad Week Three separate controlled trials published in the same week each undercut a different diet rule of thumb: avoid full-fat dairy, cut sweetness to kill cravings, treat fruit juice as sugar water. None of them are rules we chose to believe. All of them are things a food score can quietly assume anyway. -
The Nutrient The Right Answer to a Scary Headline Is Sometimes 'We Don't Know Yet' A new Wistar Institute study found a common sugar helps cancer cells escape and spread, in mice and lab dishes, not in people. What an AI nutrition coach owes a user who asks about it is a harder problem than the biology. -
The Nutrient A Federal Definition Won't Make Ultra-Processed the Right Number Days after a 200,000-person study found diet quality beats processing level at predicting disease, the FDA sent its official ultra-processed food definition to the White House. Writing a category into law doesn't fix what was wrong with asking the question in the first place. -
The Nutrient The Right Protein Number Depends on Whether You Move A 350-paper review found protein restriction mostly benefits sedentary adults, while active people and older adults preserving muscle still need more. One static protein target for every user, the industry default, is the wrong kind of simple. -
The Nutrient Ultra-Processed Was Never the Number That Mattered A 200,000-person, four-decade Harvard study found diet quality predicts heart disease and diabetes risk far better than how processed a food is. That's a problem for any app, ours included, tempted to slap a processed-food score on your plate. -
The Nutrient AI Food Photos Can't See the Butter A new NIDDK study found photo-tracking apps underestimate calories by up to 345 per meal, mostly because fat is nearly invisible in a photograph. What that means for anyone trusting a number an AI gives them. -
The Nutrient AI Just Stopped Being the Selling Point AI became a commodity in nutrition apps this year. The real moat now is trust: clinical validation, safety work, and earning the right to hold someone's health data.