Food sorting robots could soon have a strange career path: inspecting apples on a farm and helping prepare potatoes in a commercial kitchen. That is the ambition of Ficio, a young company connected to Cornell University, which is developing a common AI system for several kinds of food-handling machinery. One of its first tools evaluates apples while they are harvested; another is being designed to cut vegetables for large kitchens. The machines are at different stages of development, and the company has not demonstrated a fully autonomous food supply chain. What is interesting is the attempt to give separate pieces of equipment a shared way of recognising and handling food.
A Cornell Chronicle report published on 13 October 2026 describes the company’s work and founder Youhanna Meleka’s aim to reduce waste between farms and kitchens. FieldSorter uses computer vision to inspect fruit, while FicioPrep is a robotic food-preparation station. Both belong to a broader software idea called FoodMind. The system has been demonstrated to growers and prospective kitchen users, but the reporting distinguishes current trials from the longer-term vision. That matters because food technology demonstrations often show the best controlled examples, whereas ordinary harvests contain variation, mud, awkward lighting and materials that do not behave alike.
What food sorting robots can do at the tree
The apple machine rides through orchard rows and looks for visible defects such as bruises, punctures and damaged stems. The goal is to identify which fruit can be sold fresh and which may be more suitable for another route through the supply chain. Grading fruit at harvest could reduce the need to discover its commercial quality only after it has spent additional time in storage or travelled to another processing facility. It might also help growers make better decisions about how to handle different parts of a harvest while the fruit is still close to where it was picked.
Ficio’s system is designed to associate the grade of an apple with the tree it came from. That turns a sorting decision into information a farmer can revisit. If defects cluster in a particular area, the grower may investigate whether weather, pest pressure or handling played a role. Such data could eventually improve planning for future seasons. The useful word is ‘could’: a classification system can record patterns, but the reasons behind those patterns require agricultural expertise. An image alone cannot reveal every source of damage or every factor affecting a crop’s sale price.
The company demonstrated FieldSorter at Cornell AgriTech in New York during 2026 and told Cornell it was working with farms for further field pilots. Its own website describes FieldSorter 1 as entering customer trials during the current harvest. The same site makes distinctions between demonstrated apple picking, sorting in trials and kitchen-preparation equipment that is still described as coming soon. Those stages should not be flattened into a claim that farms and restaurants are already routinely using a completed fleet of commercial robots. The technology is real enough to be tested but remains subject to the practical work of proving reliability and economic value.
For an orchard, the economics may depend on more than detecting a bruise. Farmers need equipment that can handle harvest conditions without becoming a maintenance burden, and they must understand how the machine’s grading compares with buyers’ requirements. A false rejection could waste saleable fruit, while a missed defect might reduce customer confidence. A good sorting device therefore needs transparent performance measures and a sensible way for workers to challenge or override unusual cases. The machinery has to fit existing trailers, harvest teams and packing arrangements, not merely identify fruit in isolation.
Why the same AI is being taken into a kitchen
FicioPrep aims to recognise potatoes and prepare them into cuts suited to food service, including chips or fries. A busy catering operation may have workers spending time on repetitive preparation before cooking even begins. Automating selected cuts could change how the kitchen schedules tasks, but food preparation also involves cleanliness, safety, consistent portions and equipment that is easy to maintain. A robot that can identify a potato on a demonstration table is only at the beginning of the challenge. Commercial kitchens require repeatability, simple cleaning and predictable operation across shifts.
Cornell reports that the startup has demonstrated FicioPrep to managers at Cornell Dining while seeking feedback. The company says it wants to expand to additional vegetables and eventually add further functions. That is a development plan, not evidence that every ingredient or kitchen configuration is already supported. The cautious way to read the announcement is as an early attempt to transfer useful visual and manipulation knowledge from farm settings into food preparation. It is a promising direction precisely because the settings are so different, but the gap between prototype and standard kitchen equipment remains substantial.
The shared software is called FoodMind. Ficio describes it as independent of a particular machine design, meaning a vision and decision-making system can be adapted to different hardware. This is attractive because many robots today have to be trained for narrow, isolated tasks. A food-focused system might learn useful information about appearance, ripeness, shape and handling across machines. Yet even if visual recognition transfers, physical movements do not transfer automatically. Picking an apple from a branch involves different constraints from cutting a potato on a stable surface. A shared AI model still needs hardware-specific control and safety testing.
LiveAIWire has previously covered robots working among orchard trees, including experimental approaches to picking and pollination. The Ficio story introduces a different emphasis: finding saleable fruit and carrying information about it through the process, then applying some of the same software thinking to kitchen preparation. It also connects with earlier coverage of AI throughout the food supply chain. The problem is less glamorous than a humanoid restaurant worker, but the potential commercial question is straightforward: can machines reduce repetitive tasks and avoid avoidable losses without creating new costs or mistakes?
Why food waste is not just a robotics problem
Food can be wasted for reasons far beyond grading accuracy. Weather can damage a crop, buyers can change their specifications, transport can be delayed and perfectly edible produce may be rejected for appearance. Better information at harvest might help a grower decide where fruit should go, but it cannot create a market for everything or make perishable food last indefinitely. Likewise, an automated preparation station cannot fix a restaurant that orders too much or fails to store ingredients correctly. Technology needs to be judged by the losses it can actually influence.
A machine may also change the responsibilities of workers rather than simply removing labour. Somebody still needs to load, inspect and maintain equipment, handle exceptions, check hygiene and make commercial decisions. In some circumstances, automation could reduce exhausting repetitive cutting or sorting. In others, it could demand new skills and capital that smaller businesses struggle to afford. The distribution of benefits depends on cost, maintenance, local labour arrangements and how well the system integrates with existing practices. The company has not established those outcomes across the wider industry.
The appealing visual metaphor is a single intelligence travelling from orchard to kitchen. The operational reality is more complicated. Cameras must deal with dust and sunlight outdoors, while kitchen tools operate near sharp blades, water and food-contact surfaces. Recognition errors have different consequences in those environments. Machines may also need to work when internet access is poor. Ficio says FoodMind can run on equipment locally, an approach that could help with reliability and data handling, though the practical performance must be demonstrated under actual customer conditions.
What to look for when the trials become commercial
The decisive evidence will not be a promotional image of a machine next to a bin of fruit. It will be independently understandable figures about grading accuracy, throughput, downtime, cleaning time, labour requirements and whether losses actually fall. For the kitchen system, proof would include safe operation with varied ingredients and reliable results over sustained use. For the orchard system, results across different fruit varieties and difficult harvest conditions will be more informative than a short demonstration. Buyers also need to know whether savings justify purchasing, servicing and supervising the equipment.
LiveAIWire has covered the wider effort to bring AI into farming and food production. Ficio offers a compact illustration of that ambition, but the company’s own status labels are a useful restraint on grand claims. One application is in trials, another has been demonstrated, and more ideas remain to be built. If its common software genuinely helps different machines adapt to food-handling work, that would be an important engineering achievement. The interesting story today is a startup testing that possibility, from the apple tree to the chopping board, while the everyday economics are still being worked out.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.