CogniCuisine is not about asking an AI for a recipe and cooking what comes back. It is about bringing human expertise, culinary instinct, and critical thinking to a collaboration where AI earns its place at the table by doing something specific and useful — and where the cook remains in charge of every decision.
This is the full account of how a single recipe moved from a first draft through chemistry analysis, cultural grounding, stove testing, and real-time correction to a dish that worked. It is also an honest account of what AI got right, what it got wrong, and where human judgment was the only thing that closed the gap.
It is also, in this case, a demonstration that the methodology applies even when the AI being interrogated is the same one that wrote the original draft.
Stage One: The First Draft
This recipe began with Claude. In February 2026, Meyers-Lussier asked Claude to generate ten original unconventional breakfast concepts for the “Unconventional Breakfasts” episode of A Seasoned Friendship. Labneh Pancakes with Sour Cherry Compote & Za’atar Butter was number eight on that list.
Claude then searched the existing recipe landscape to verify originality. Labneh had appeared in biscuits, in buns with za’atar, and as a topping on pancakes. But labneh folded into the pancake batter itself, combined with sour cherry compote and za’atar compound butter? No published recipe matched that combination. The concept was cleared as original.
Claude wrote the initial recipe. The structure was sound — labneh folded into the batter, sour cherry compote, za’atar compound butter melting into the stack. Levantine ingredients in an American breakfast format. The concept had genuine merit.
The chemistry had not been tested.
That is not unusual for a first draft. Generating a plausible recipe is not the same as pressure-testing it. The CogniCuisine methodology treats any first draft — including its own — as a starting point for interrogation, not a finished product. The recipe went to Qwen for chemistry review, with specific attention directed to the compote.
Stage Two: Chemistry Analysis with Qwen
Qwen was brought in specifically for its chemistry reasoning. The question put to it was direct: review this recipe for properly built chemistry and flavor profiles, with particular attention to the compote.
Qwen identified three structural problems in Claude’s draft before anything reached the stove.
Problem One: The Compote Would Not Set
Claude’s original recipe specified one teaspoon of cornstarch for two cups of sour cherries — insufficient by a wide margin. Cornstarch requires roughly one tablespoon per cup of liquid to create a proper gel. Sour cherries release significant liquid during cooking. The original recipe would have produced a thin syrup, not a compote with enough body to sit on a pancake stack.
When Meyers-Lussier clarified that jarred unsweetened cherries, not frozen, were the ingredient on hand, Qwen recalculated. Jarred cherries carry less free liquid than frozen. The revised specification was one and a half teaspoons of cornstarch. That was closer. It was still not correct — but that discovery belonged to the stove.
Problem Two: The Acid Load
Claude’s original recipe combined four overlapping acid sources: labneh (lactic acid), sour cherries (malic acid), lemon juice (citric acid), and za’atar, which contains sumac — itself a significant source of tartness. Qwen identified this as a structural problem. Four acid sources without sufficient sugar or fat to buffer them produces a sharp, astringent finish rather than the bright, balanced profile the recipe intended.
The fix: reduce lemon juice from one tablespoon to one teaspoon, and increase the compote sugar from a quarter cup to a third of a cup.
Problem Three: The Za’atar Butter Required Chewing
Compound butter is meant to melt. Za’atar contains dried herb fragments, toasted sesame seeds, and ground sumac — none of which dissolve in cold fat. Mixed into softened butter and refrigerated as Claude’s recipe specified, those particles stay distinct. The result is a butter that requires chewing to extract its flavor, which is counterintuitive for something intended to perfume a pancake as it melts.
Qwen identified the problem and the chemistry behind it. The solution — blooming — came from Meyers-Lussier.
Stage Three: The Cook Directs
This is where the methodology becomes visible as a methodology rather than a recipe generation service.
The compote aromatics. After the chemistry issues were resolved, Meyers-Lussier asked whether the compote needed a flavor element — a spice, an herb, or an oil — to complete it. Qwen presented options. Meyers-Lussier then set the constraint: keep the recipe ethnically true to its Levantine and Turkish roots. Within that constraint, Qwen identified cinnamon as the authentic spice for a Turkish-style sour cherry compote (vişne reçeli) and orange blossom water as a classically Levantine aromatic finish. Meyers-Lussier selected both.
The blooming method. Meyers-Lussier asked whether the za’atar could be pre-treated — bloomed in the butter before refrigerating — to solve the texture problem. Qwen confirmed the chemistry: blooming releases volatile oils into the fat, softens herb particles, and produces a smoother mouthfeel. Meyers-Lussier originated the approach. Qwen explained why it would work.
The raw sumac. Meyers-Lussier proposed adding raw sumac to the butter after it cooled, rather than blooming it with the za’atar. Sumac is a fine powder, not a dried herb. It disperses without heat. Adding it off-heat preserves its volatile citric aromatics — the bright, clean tang that distinguishes it from the mellower flavors the bloomed za’atar provides. Qwen confirmed the chemistry. The decision was Meyers-Lussier’s.
Stage Four: The Stove
This is where AI — all of it, including the AI that wrote the original draft — reaches its limit and the cook takes over completely.
The compote cornstarch. The compote was built to Qwen’s corrected specification — one and a half teaspoons of cornstarch. At the stove, it was clear that was still insufficient. The specific liquid content of a particular brand of jarred cherries is something no AI system can know in advance. Meyers-Lussier doubled the cornstarch to three teaspoons. The compote set correctly.
The orange blossom water. The quarter teaspoon Qwen specified was confirmed correct from the plate. The aroma arrived before the taste — orange blossom water is a volatile essence, and the nose catches it first. It settled as an overtone to the cherry rather than a dominant flavor. That quarter teaspoon is the ceiling. More would have overwhelmed the compote.
The batter. When wet ingredients met dry, the batter puffed visibly — the labneh and baking soda reacting immediately. One tablespoon of milk brought it back to a thick, pourable consistency. The finished pancakes were light, airy, nearly half an inch thick, tender, and moist.
What This Recipe Demonstrates
The CogniCuisine methodology does not exempt any AI from scrutiny — including the one that generated the concept. Claude produced a recipe with real structural flaws. Those flaws were caught because the methodology requires that every draft, regardless of source, be pressure-tested before it reaches the stove.
Qwen’s chemistry analysis eliminated the most predictable failures before the first pan was heated. It narrowed the variables significantly. But it could not know the liquid content of a specific jarred cherry brand. It could not smell the orange blossom water before it was added. It could not feel the batter thicken under a spoon. Those calibrations belong to the cook.
The creative and culinary decisions in this recipe — the cultural constraint, the blooming method, the raw sumac, the stove corrections — were all Meyers-Lussier’s. Claude generated the concept and the first draft. Qwen provided chemistry analysis when asked. The cook directed the work from the first prompt to the final plate.
That is the division of labor CogniCuisine is built on.
The Finished Recipe at a Glance
Labneh Pancakes with Sour Cherry Compote & Za’atar Butter
Serves 4 — about 12 pancakes
The pancake batter uses labneh for a tender, lightly tangy crumb. The compote is built in the Turkish vişne reçeli tradition — sour cherries, sugar, and cinnamon — finished off-heat with orange blossom water. The za’atar butter is bloomed in melted butter to smooth the herb particles, then finished with raw sumac added off-heat to preserve its bright citrusy aromatics.
Cornstarch in the compote: three teaspoons, not the one Claude’s original draft specified and not the one and a half that Qwen’s chemistry analysis predicted. The stove had the final word.
Before & After: What Changed and Why
The table below shows Claude’s original draft alongside the final tested recipe. Every change is annotated with the reason — whether it came from Qwen’s chemistry analysis, Meyers-Lussier’s direction, or the stove itself.
| Element | Claude’s Original Draft | Final Tested Recipe | Source of Change |
|---|---|---|---|
| Compote cornstarch | 1 teaspoon | 3 teaspoons | Qwen identified insufficient thickening; stove doubled it further |
| Compote sugar | ¼ cup | ⅓ cup | Qwen: four acid sources required more buffer |
| Lemon juice | 1 tablespoon | 1 teaspoon | Qwen: acid load too high with labneh, cherries, and sumac already present |
| Compote spice | None | 1 small cinnamon stick | Meyers-Lussier: cultural authenticity constraint; Qwen identified Turkish vişne reçeli tradition |
| Orange blossom water | Not present | ¼ teaspoon, off-heat | Meyers-Lussier selected from Qwen’s Levantine options; confirmed correct at the plate |
| Za’atar butter method | Mix into softened cold butter | Bloom in melted butter, cool, re-emulsify | Meyers-Lussier proposed blooming; Qwen confirmed the chemistry |
| Sumac in butter | Bloomed with za’atar (implied) | Raw sumac added off-heat after cooling | Meyers-Lussier: preserve volatile citric aromatics; Qwen confirmed |
| Batter milk | ¾ cup | ¾ cup plus 1 tablespoon as needed | Stove: labneh-baking soda reaction thickened batter beyond pourable consistency |
The pattern is consistent with what the CogniCuisine methodology predicts: AI generates a plausible structure, chemistry analysis catches the predictable failures, human judgment sets the creative and cultural parameters, and the stove resolves what neither could know in advance.
For researchers and professionals: The full Cook Mode formatted recipe document, including all methodology notes, is available for download below.