How an AI support tool turned sceptical chefs into believers

Most upmarket restaurant menus begin life as scribbles on scrap paper. Usually drafted by an executive chef, in collaboration with other senior members of the kitchen brigade, these epicurean imaginings are mulled over and allowed to marinate. 

Ideas are proposed, revised, discarded and revived. Ingredients shift. Techniques are debated. Over time, the scribbles cohere into recipes that are cooked, tested, tasted and adjusted accordingly.

Sketches of potential plating are put forward. Balance and flow — within each plate, from one plate to the next and across courses — are considered and refined. Costings are calculated. Suppliers are sourced. Mise en place, kitchen preparation, lists are devised. Only then does the menu move from concept to final form, written, proofread, and ready to be printed or published.

Such menus are generally revised regularly — at least seasonally, sometimes more often. This means that this happens at least four times a year. Almost always, the intention is that each iteration will be recognisably of the house style but still distinct from previous offerings. Balancing consistency with change, tradition with innovation. A pinch of novelty, a dash of classic technique, a splash of surprise folded into broadly familiar flavours. 

Diners return for what they loved before, but also to be delighted anew. They want to see their favourite chefs responding to shifting food trends and evolving tastes, while remaining true to their culinary voice.

To repeat this process, over and over, season after season, year after year can be exhilarating, but it is also exhausting. Especially if the menu creator is also managing staff dynamics, health and safety protocols and the ambient pressure of economic uncertainty. Faced with such demands, very few chefs would reject well informed, always available support — especially from an unemotional entity that doesn’t feel offended if its suggestions are rejected or reframed.

Enter the Culinary Creative Navigant (CCN). This Generative AI tool has been designed to support chefs through the menu development cycle. Users begin by entering their personal and/or restaurant profile. From there, a guided culinary conversation unfolds, exploring technical, conceptual and sensory preferences. Together, the chef and the CCN can co-create each element. 

It doesn’t replace the chef but rather acts as an educated subordinate, always ready with a suggestion or to act as a supportive sounding board.

While the creative process of flavour architecture, plating aesthetics, between course balance and beverage pairing is under way, the AI Navigant simultaneously handles logistics: compiling equipment and ingredient lists, calculating costings and organising production schedules.

It generates detailed descriptions of every menu item and articulates the chef’s intent for staff training, both back and front of house. It collates answers to frequently asked questions from diners so that waiters always know whether there is gluten in the sauce, or the origins of an ingredient. It can even draft social media content drawn from the menu’s narrative.

Part sous chef, part kitchen manager with elements of maître d’hôtel and social media content creator, because the CCN is tailored to the chef and restaurant’s unique identity, every interaction is singular. No two  interactions are ever the same. It’s not just a tool — it’s a creative companion, designed to help chefs sustain their vision while navigating the practical demands of modern hospitality.

Which sounds wonderful — in a space age, theoretical kind of a way — but does it work in practice, and if so are chefs ready and willing to accept such a sidekick?

As unlikely as it sounds, I have an uber-intelligent tech-bro buddy who has created just such a gastronomic gizmo. He asked me to use my contacts in the restaurant world to create a pilot project exploring the role of Generative AI in culinary creativity. He chose me because he knows that I hate AI. I am simultaneously irritated, unnerved and threatened by its existence. He figured that I was unlikely to show his invention to chefs who would automatically admire it.

And he was right. The first chef I showed it to refused to be a part of the pilot project, saying: “Why would I want to give that machine any role in my creativity? That’s my private space, my process and my favourite part of my job.” 

I had more luck at the Prue Leith Culinary Institute, Centurion. Starting in May 2025, they graciously agreed to test the CCN while designing the winter menu for the institute’s eponymous, luxurious and deeply delicious modern South African restaurant.

The chef-lecturers were given an afternoon of basic prompt training and then left alone to interact with their new kitchen companion. During the initial education session, most were curious but wary. Some considered the use of AI to be a form of cheating. Several feared that they lacked the computing skills to make use of the model. Everyone was concerned that it might interfere with the hitherto profoundly personal (and intensely satisfying) act of menu design.

The unease was entirely understandable. Professional chefs spend years mastering their discipline through education and repetition. Over time they acquire the kind of super-skilled technical muscle memory that is often literally seared into them through contact with blistering oil and/or molten sugar. The idea that a chatbot could contribute meaningfully to their ancient art seemed not just unlikely, but actively insulting. Which is why what happened next was so surprising...

After a sceptical start, enthusiasm, innovation and some deeply delicious recipes emerged. When I went back in late July to interview my study subjects, pastry chef Lesego Mabale admitted: “The first attempts were not great. We didn’t really understand its capabilities, so we were treating it like Google — typing in a particular preformed concept or asking for a specific recipe — for instance ‘cocktail lamb bunny chow’ and getting back a result that we either accepted or rejected. We weren’t engaging with it or refining its ideas.”

The breakthrough came when they stopped thinking of CCN as a search engine and started treating it as a very well-read but literal-minded apprentice. 

As hot kitchen chef Cori-Li Vermaak pointed out: “You have to spell it out like a grade schooler. If you tell it something you get exactly that. You need to describe what you hope to achieve, carefully and in detail. You have to clearly set out your aims — every detail of what you want. I want to include these sorts of tastes and textures. I want it to show environmental sustainability or national pride, or whatever it is. The more detail, the better the result. It is not a mind reader. If you are clear, the results are what you want. The process pushes you to think about what you really want, why you want that, what that will achieve.”

The shift in approach yielded results. Cori-Li’s hot kitchen colleague chef Quinton Els observed: “The process is not about being given an idea and accepting it without thinking. It’s a conversation. What ultimately comes out is a hybrid dish. There is nothing on the new menu that was just taken as it came out of the machine. We added on. We took out. After each suggestion we refined and modified. The prompting is like a kind of thinking out loud. There was real innovation that came from the process, but it came from collaboration, rather than just waiting for it to come fully formed from (someone or something) else.”

Lesego added that “when you test an idea that has come out of that collaborative process you then find all sorts of things you want to modify. For instance, we were working on a trio of apple sorbets. We made the recipes and decided that the textures between them were too similar. We wanted some of them to be silky smooth but others to have that slightly textured quality that eating an apple can have. So then we added back some of the pulp that the recipe had told us to strain out initially.”

Cold kitchen chef Natasha Coradi-Sutton liked its ability to rapidly provide multiple options: “It was nice to have lots of ideas. When asked about a particular plate, many suggestions would come out. We seldom used everything — we mixed and matched. We would look and see if this element might work well on a different dish. It just made for wider possibilities.”

It is all very well to collaborate on exciting ideas, but did the recipes that accompanied those ideas actually work? Did the cooking times suggested burn the biscuits? Was there sufficient seasoning in the sauces? Were the unusual flavour combinations it came up with compatible? 

Going into the pilot project, neither those who had created the CCN nor the chefs were sure that it could write a usable recipe. The good news is that it can. 

Chef Christalle Soules said: “In the pastry department we are all about accuracy, and the recipes really worked. First time. Every time. That is not to say we didn’t modify them to suit our style, but the actual recipes could be used without problems.” 

The hot kitchen team initially had minor issues with meat cooking times. These predominantly pertained to variations in fat content and were fixed by providing the AI tool with additional information.   

In addition to the CCN’s creative capacities and ability to write a recipe that produced good food, it also took on the schlepp work. The chef-lecturers are perfectly capable of adapting recipes and menus to suit cultural and medical dietary restrictions, but it is undoubtedly a time-consuming task. It took the CCN mere seconds to offer up a South African heritage-inspired afternoon tea menu that was also suitable for a diabetic.

This freed up the chefs to do more of what they love. The chefs felt that AI supported their creativity by taking on the routine administrative tasks — scaling recipes, calculating nutritional information, generating shopping lists, writing menu descriptions. 

Head chef Anel Louw reported spending 40% less time on paperwork, using those hours to work with students and develop new dishes. As Lesego observed: “There was a fear that this experience would make us lazy, but it didn’t. Efficient yes, but lazy no.”

At the initial training the chef-lecturers referred to the CCN as “it” but by the time we returned in late July, their accounts were of what “she” had done. Some of the staff were calling her Sonia. This anthropomorphising suggests that the technology had begun to cross the threshold from threatening tool to trusted team member. Charmingly, the tech-bro inventor assumed that Sonia was an acronym, but the chefs assured him that it was “just a name that a cheeky tannie would have”.

The team’s growing fondness for Sonia did not blind them to her limitations. Sonia performed impressively in generating novel concepts and remixing global trends, but she had some limitations when it came to incorporating southern African indigenous ingredients, methods and culinary cultures. 

When asked to include local elements she was very focused on certain ingredients and recipes — as Anel observed: “Sonia is a big fan of chakalaka. In the hot kitchen she likes to suggest it with everything. You have to tell her ‘no’.” 

The model’s blind spots seem less a failure of design than a reflection of the digital archive it draws from — an internet that remains thin on African culinary knowledge, underrepresenting the aesthetics, techniques and philosophies that shape modern African restaurant kitchens. The developers are currently working to expand the database from which Sonia draws.

Plating was also not Sonia’s strong suit. The chefs found her attempts at visual presentation dated and dull. Perhaps her suggestions would have worked for lesser chefs, but those at the Prue Leith Culinary Institute consistently create a level of beauty and sophistication that was beyond Sonia’s current capacities. Her ideas were generic, and when this was pointed out to her she seemed incapable of upping her game. When the chefs told Sonia that her sauce dots did not do it for them, she simply moved the dots to the other side of the plate and represented the picture.

Sonia initially used retail prices — sourced from supermarket websites — thereby creating astronomical food costs, but that was rapidly resolved by providing her with wholesale supplier data (and an explicit prompt not to look to supermarkets for prices). 

With portion sizes she also needed clear instructions as to what were culturally appropriate amounts. She was unaware that the Prue Leith restaurant’s Centurion clientele is culturally predisposed to favour slightly larger portion sizes (especially when it comes to meat) than those found in the international recommendations.

She seemed to prefer smooth textures and had to be reminded to consider texture. As Lesego said: “When I create recipes, I think them through in my mind. I am tasting in my mind, imagining the sensation of the mouthfeel, hearing the crunch of a bite because I am a real person and I have had those experiences. She is not a person, so she needs to be schooled.”

Sonia was also quite wasteful — because no one had told her not to be.

Christalle said: “The recipes tasted good. The flavour was definitely there but she is quick to throw stuff away. Useful stuff. Like when we were making the biodiversity sorbets, the recipe given says to make a purée, drain it, use the juices and then you chuck out the pulp! And we then thought, okay, this is going to be a waste. Let’s try to make fruit leather. And so we made that from the pulp. 

“And it makes it beautifully. You don’t have to add anything extra to it. You just literally spread it very thin, put it under the pass, dry it out, and that’s it. Much better than just chucking that in the bins.” 

Chef Cori-Li agreed: “She suggested a radish pesto, which was an idea that we went with but we chose to blanch the leaves and put those in the blender too. She was going to just throw them away. Once she was informed that at Prue Leith we have a zero-waste approach she quickly caught on. 

“That experience led me to write up a list of prompts that I keep next to my desk. It is all the things that I need to tell Sonia when I start talking to her. I tell her that we have a focus on indigenous ingredients, that we are a zero-waste kitchen and so on before I start.”

As Adele Stiehler-van der Westhuizen, Managing Director of the Prue Leith Culinary Institute, said: “When we first started, the recipes produced were consistently formatted for home cooking rather than professional kitchens. Professional chefs work at subdivided stations, prepare in bulk, and follow standardised mise en place procedures.

“Initially, recipe methods had to be restructured for restaurant service, but that was about us not telling Sonia what we needed from her. We needed to actually engage. To be very clear at all times and tell her about how we like to work. To tell her when she wasn’t giving us what we wanted — not just that’s not correct, but why it was wrong so she could learn. We learnt that skill.” 

In AI jargon that is referred to as “prompt engineering” — this kitchen brigade called it “teaching Sonia to speak chef”.

The proof of the pudding is in the eating. Diners are offered a choice of six starters, six main courses and six desserts. Whether you want umleqwa chicken and rooibos tea consommé, pork belly with a baobab and nartjie glaze on chakalaka potato purée — yes, Sonia got her way in one dish — or a sorghum ravioli, the savoury offerings are exquisite.

Desserts are as delicious as they are beautiful. The aforementioned apple biodiversity sorbet trio (featuring Golden Delicious, Pink Lady and Granny Smith apple ices with walnuts, fruit leather and apple tartare) competes for customer attention with the boozy bliss of “polisie koffie” crêpes with coffee ice cream, milk skin shards, brandy butter and Ouma-style rusks.

Prue Leith Culinary Institute is not only a restaurant, but also a chef school. The implications of AI for culinary education are considerable. The chef-lecturers felt that junior students should first learn to create menus the old fashioned way before being introduced to Sonia because, as Anel observed: “They would be very easily impressed by the dishes that are brought forward — think it’s very cool and not take it further. They would probably not interrogate that content. I am not sure they would judge the output and think ‘How can I develop this?’ And that might stop the growth of their creativity. I think it’s ideal for students who have got beyond that stage.”

Interaction with Sonia has altered the way that the chef-lecturers regard AI. They are more willing to see other aspects of AI as a study aid. Previously they saw it as a way that students might cheat; they are now seeing it as the equivalent of a supportive librarian guiding a student to the correct study material and an ever-present teaching aid getting the students to understand the process by which knowledge and skill are acquired.

Adele said that “we must not try to figure out whether the students use AI, we must assume that they will use AI. Our question should be how can we test their ability to use it as a culinary tool? Do they understand that how to ask the right question is really the important thing? And do they have the knowledge to assess the value of the information that they are given? Where can they take the information next?”

Henceforth Prue Leith students will not need to use AI furtively, but will rather be required to submit their entire AI conversation, showing how they refined prompts, challenged suggestions, and built upon machine-generated foundations. As Adele explains: “We’re not teaching them to use AI as a crutch. We’re teaching them to use it as a springboard. The difference matters.”

The chef-lecturers were unanimous: AI won’t replace them or their students any more than electric mixers rendered bakers obsolete. Those who master these tools will have significant advantages. It is not about choosing between tradition and innovation. It’s about finding the sweet spot where ancient knowledge meets modern tools. 

Culture cannot be reduced to data points or tradition to algorithms, but technology can amplify the human capacity for creativity. Chefs aren’t being swopped out — they’re being empowered with new resources, but human skill and soul is irreplaceable.

The CCN is currently still in closed beta testing, and not yet available to the public, but chefs interested in early access should contact me. I am looking for more gourmet guinea pigs. DM