Kitchen Chemistry Companion

The problem

I do not enjoy cooking. I enjoy delicious food, and cooking is the maintenance task inconveniently positioned between me and continued existence.

The conventional recipe workflow asks me to choose future meals, check what I have, shop for exact ingredients, remember the plan and cook everything before the perishables become compost. That is a lot of planning, memory and timely follow-through for one meal. I cook from the ingredients already in the kitchen because that removes much of the executive work.

That adaptation creates a different problem. A recipe may be almost workable, apart from the one ingredient I do not have. Without knowing what that ingredient does, I cannot tell whether I can omit it, replace it or accidentally turn dinner into either soup or masonry.

The origin

Lessons in Chemistry is Bonnie Garmus’s novel, later adapted for television, about Elizabeth Zott: a chemist who becomes the host of a television cooking programme and treats cooking as applied chemistry. She respects her audience enough to explain what is happening rather than issuing instructions from the countertop.

Watching the series gave me the missing diagnosis. Recipes told me what to do, but I lacked the causal model that would explain why they worked. Chemistry was one of my least favourite school subjects, the periodic table has not wandered back voluntarily, and decades have passed since anyone expected me to remember it.

I am interested in the outcome and in enough understanding to produce it. I am not interested in mastering the entire underlying discipline before I can make dinner. In this domain, I learn best through an immediate problem, an explanation I can use and another attempt.

From barrier to design response

Observed problem	Underlying barrier	Design response
Planning meals and shopping for exact recipes creates too much executive work	The workflow depends on future planning, memory and timely follow-through	Support ingredient-first cooking with what is already available
One missing ingredient can stop the whole attempt	A name-based substitute list does not explain the ingredient’s function	Identify the chemical and structural role before suggesting an adaptation
Recipe instructions do not transfer to the next unfamiliar situation	I receive a procedure without the causal model behind it	Explain the relevant mechanism while helping with the immediate meal
A failure may become visible only when it is difficult to reverse	Splitting, curdling, collapse and overhydration have earlier warning conditions	Predict likely failure modes and flag them before the irreversible step
Studying culinary chemistry in the abstract will not happen	I become interested when the knowledge solves a live problem	Teach in small, usable pieces during practice so understanding accumulates
Too many questions can make the support as burdensome as the recipe	Every interruption adds another point at which momentum can disappear	Ask only when the answer materially changes the likely outcome

The wider idea

Kitchen Chemistry Companion brings specialist reasoning into a low-stakes personal activity where I would never have hired someone and where mastering the complete discipline would be wildly disproportionate.

The level of expertise you need should be proportionate to the outcome, the stakes and your responsibility to other people. Cooking my own dinner calls for a different arrangement from developing a commercial food product, advising someone with a medical condition or preparing food in a regulated setting. Those contexts need qualified expertise, testing and accountability.

The design question is: what outcome do I want, which part should remain mine and what missing capability would extend my range? The answer may involve learning the discipline, hiring a qualified expert or designing a bounded collaborator. The stakes determine which.

This is the same logic I develop in Designing Your Co-Intelligence: understand how you already work, find where friction appears and decide which capabilities deserve a place around you.

What I built

Kitchen Chemistry Companion started as a custom GPT and later became a private ChatGPT plugin. I designed its culinary-science model using references including Mouthfeel by Ole G. Mouritsen and Klavs Styrbæk, Culinary Reactions by Simon Quellen Field and Chemistry of Spices edited by V. A. Parthasarathy, B. Chempakam and T. J. Zachariah.

The collaborator reasons about ingredient systems: water and moisture, fats and emulsions, proteins, starches, sugars, acids, hydrocolloids and binders, and volatile compounds. It looks at what heat, mixing, timing and sequence are likely to do to them.

A substitution follows function rather than resemblance. The companion identifies what the missing ingredient contributes, checks which available ingredients can supply enough of that function and explains the trade-off. It also flags likely failure modes such as splitting, curdling, gumminess, rubberiness, collapse, dryness and accidental educational lumps.

How responsibilities are divided

The companion contributes	I retain
Ingredient roles, mechanisms and functional substitutions	The ingredients I want to use and the outcome I am trying to create
Predictions about texture, stability, heat and likely failure	Taste, sensory judgement and decisions during cooking
A targeted explanation when the mechanism will help	Whether I want the explanation now or only the immediate adjustment
Questions whose answers change the recommendation	Responsibility for food safety, dietary needs and the final meal

Kitchen Chemistry Companion was not designed to make me follow recipes more reliably. It was designed to make my existing ingredient-led way of cooking more resilient.

Learning while doing

The collaborator solves an immediate problem and leaves a little more understanding behind. A failed emulsion becomes an explanation of water, fat and stabilisation. A dense cake becomes a lesson in structure, gas and mixing. The next time a related problem appears, I have more of the model myself.

The learning follows use rather than a curriculum. Knowledge arrives with a reason to care, an example I can inspect and a chance to apply it before it evaporates.

Make your own

I created a reusable workflow for designing a kitchen collaborator around your cooking habits, constraints and learning preferences. It helps you diagnose the actual workflow, identify the missing capability, choose appropriate knowledge sources, divide responsibilities, set safety boundaries and test the result against realistic situations.

The workflow produces a custom instruction set. It does not reproduce my private plugin or assume that anyone else shares my particular relationship with meal planning, chemistry or educational lumps.

• Design your own Kitchen Chemistry Companion Markdown · guided workflow and copy-ready instruction template · Plain text version

What this project demonstrates

Capability	Evidence in this project
Behavioural diagnosis	I traced the cooking problem past motivation to the executive demands of recipe selection, stock-checking, shopping, remembering and timely use.
AI collaborator design	The specification gives the companion a bounded role: supply culinary-science reasoning while leaving taste, improvisation and final judgement with me.
Workflow design	The collaborator supports ingredient-first cooking and limits questions to information that changes the recommendation.
Knowledge translation	The model turns culinary chemistry into functional ingredient roles, predicted interactions and timely explanations.
Embedded learning	Each intervention can solve the meal-level problem while adding a causal explanation that transfers to later cooking.
Risk calibration	The design distinguishes low-stakes personal improvisation from commercial, medical and regulated contexts that require qualified expertise.