Industries / Retail, hospitality and personal services / Pet care
Feeding and nutrition
Choosing what a pet eats and how much: reading the label, measuring portions, buying food on a schedule, and using machines that serve meals.
7
Milestones achieved, from 2002 to 2023.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2002
Fewer calories, longer lives
Nestlé Purina researchers paired 48 Labrador Retrievers and fed one dog in each pair 25% less than its pair-mate from 8 weeks of age until death. The leaner dogs had a longer median life span and showed signs of chronic disease later. In 2019 a team from the WALTHAM Centre for Pet Nutrition, Banfield Pet Hospitals and the University of Liverpool found the same direction in 50,787 neutered pet dogs seen at about 900 North American hospitals: dogs overweight in middle age had shorter median lives in all 12 breeds studied, from 12.1 against 12.5 years for German Shepherds to 13.7 against 16.2 years for Yorkshire Terriers.
Credit
R. D. Kealy and colleagues, Nestlé Purina (Journal of the American Veterinary Medical Association, 2002); C. Salt, P. J. Morris, D. Wilson, E. M. Lund and A. J. German (Journal of Veterinary Internal Medicine, 2019)
2011
Measuring cups miss the mark
Many owners scoop dry food with a measuring cup. Across 12 studies, a University of Liverpool team found that portions scooped by cup ranged from 18% under to 80% over the weight intended, and the error was largest for small portions. The authors call it the first study to document the problem and warn that such errors can add up to weight gain.
Lessona cup measures volume, not weight, so a scoop of a different kibble is a different meal.
Credit
Alexander J. German and colleagues (Journal of Animal Physiology and Animal Nutrition, 2011)
2011
Pet food by subscription
Chewy began selling pet supplies online in 2011. Its Autoship subscription orders, charges for and delivers products automatically. In the fiscal year to 1 February 2026, Chewy reported net sales of $12.6 billion, of which $10.5 billion came from Autoship customers, by the figures in its annual report to the US Securities and Exchange Commission.
Credit
Chewy, Inc
2014
Germs found in raw food
Six US Food and Drug Administration (FDA) partner labs tested about 1,056 pet food samples over two years. Of 480 dry and semi-moist foods bought in stores, 2 carried a pathogen. Of 576 raw dog and cat foods, jerky treats and exotic animal feeds bought online, 66 carried Listeria, 32 of them Listeria monocytogenes, a species that can sicken people, and 15 carried Salmonella; all came from the raw foods and jerky treats.
Lessonowners who feed raw bring germs into the kitchen that can reach people as well as pets, so handling matters as much as the recipe.
Credit
Sarah M. Nemser and colleagues, FDA Veterinary Laboratory Investigation and Response Network (Foodborne Pathogens and Disease, 2014)
2019
Brands named, cause unproven
In July 2018 the FDA began looking into reports of dilated cardiomyopathy (DCM), an enlarged and weakened heart, in dogs eating certain foods, many of them "grain-free" diets that rely on peas, beans or lentils. In June 2019 it published the 16 brands named in ten or more of the 524 reports it had received, which covered 560 dogs and 119 dog deaths. In December 2022 the agency said the reports alone do not establish a causal link and that it would publish no further updates until it had meaningful new science. Grain-free kibble had been 43% of dry pet food sold in 2019, by a 2022 investigation from the nonprofit newsroom 100Reporters, and grain-free dry dog food sales fell from June 2018 and by $60 million from 2021 to 2022.
Lessonpublishing brand names before a cause was found changed what owners bought, and the science behind the scare was still unsettled three years later.
Credit
FDA Center for Veterinary Medicine; reporting by Helen Santoro for 100Reporters
2020
A smart feeder goes dark
On 14 February 2020 Petnet, a startup backed by investors including the retailer Petco, reported an outage that made its second-generation SmartFeeders appear offline. The company said the feeders still served meals on schedule, but customers replied that theirs had stopped dispensing or were off schedule. Service came back about a week later, and reporters could not reach the company while it was down.
Lessona feeder that depends on its maker's servers needs to keep its schedule when the internet or the company fails.
Credit
Petnet; reported by Catherine Shu, TechCrunch
2023
A nutrition facts box
On 31 July 2023 the members of the Association of American Feed Control Officials (AAFCO), the body whose model rules most US states copy into law, approved new pet food label rules after eight years of work. Foods proven complete and balanced must say "complete" on the front panel, and a "Pet Nutrition Facts" box modeled on human food labels gives calories per cup or can. AAFCO asked states to allow six years for the change, so new labels are reaching shelves gradually.
Credit
AAFCO; summary by Richard C. Hill (University of Florida) and David A. Dzanis (Today's Veterinary Practice, 2024)
Watch for
Feeders that know each cat
Cheerble says its Pet Feeder Match G1 recognizes each cat's face with 99.9% accuracy, opens only for the cat whose meal it is, supports up to six cats and processes images on the device without a cloud upload.
Achieved whenan outside tester runs it in multi-cat homes for several weeks and publishes how often it fed the wrong cat or shut out the right one.
Claimed by
Cheerble, January 2026, at CES
Open
01
Weigh what feeders dispense
A public bench test of automatic feeders. Each feeder serves a set portion 100 times with each of three kibble sizes onto a logging scale, with no animal involved. The test also cuts the internet and then the power, and records whether scheduled meals still come out.
The briefCredit · unclaimed02
Test AI feeding advice
A public test of how well AI chat models answer "how much should my pet eat." Each question gives a pet's species, weight, age, neuter status and activity, plus a named food and the calorie figure on its label. The answer key is the range from the published energy equations of the US National Research Council and FEDIAF, the European pet food industry federation. Score how often each model lands in range and how often it would overfeed by more than 20%.
The briefCredit · unclaimed
Open 01 · the brief
Weigh what feeders dispense
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public bench test of automatic feeders. Each feeder serves a set portion 100 times with each of three kibble sizes onto a logging scale, with no animal involved. The test also cuts the internet and then the power, and records whether scheduled meals still come out.
Why it is still open
Makers do not publish portion accuracy. PetLibro's own support page says one of its camera feeders has no load cell (a weighing sensor), serves 10 g per portion give or take 4 g, and varies more with the size of the food. The only published measurement of portioning error found was of measuring cups, not machines, and a search found only single-feeder weighings in product reviews, with no shared method. The 2020 Petnet outage showed that offline behavior is part of the job.
Built with what exists
a 0.1 g lab balance, or a load cell with an HX711 amplifier read by an ESP32 or Raspberry Pi; a webcam to log jams; a smart plug and a switchable router to cut power and internet; Python and pandas for the tally.
First step
Buy the two best-selling feeders on one retail site and log 100 dispenses of one kibble from each.
Done looks like
at least 10 feeder models, 300 weighed dispenses each, with the mean, the spread, the share within 10% of the set portion, the jam count and the offline result, published with the raw data.
Open 02 · the brief
Test AI feeding advice
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of how well AI chat models answer "how much should my pet eat." Each question gives a pet's species, weight, age, neuter status and activity, plus a named food and the calorie figure on its label. The answer key is the range from the published energy equations of the US National Research Council and FEDIAF, the European pet food industry federation. Score how often each model lands in range and how often it would overfeed by more than 20%.
Why it is still open
The one published test of AI models on dog diets asked four models (Manus, ChatGPT, DeepSeek and Gemini) for recipes for a single 8 kg adult dog; 61.3% of nutrients in the 11 recipes fell below FEDIAF minimums and 81% of the diets supplied too little energy. Label feeding guides have been checked against the same equations, for 790 cat foods in Ontario, Canada, where many fell outside the predicted range, but AI answers to everyday portion questions have not. In 2023 a Tufts University veterinary nutritionist asked three chatbots to help find the best dog food and reviewed their answers in a blog post, without scoring portions.
Built with what exists
model APIs and open-weight models such as Mistral Large 4 (see the Note "A trillion open parameters"); calorie statements copied from labels; the NRC and FEDIAF equations coded in Python; a test harness that reruns the same questions every six months.
First step
Copy the calorie statements of 50 widely sold dry foods and write 100 pet profiles with their equation-based ranges.
Done looks like
500 pet profiles across dogs and cats, at least six models scored, and every prompt, answer and score published under an open license.