A Recipe for Thought:Why Structured Frameworks and Generative AI Are the Future of Thinking
By Anthea Roberts, Nick Lothian, Dimitri Glazkov and a.r. Routh
Picture this: a beautifully stocked kitchen—gleaming countertops, the finest ingredients, and state-of-the-art equipment lined up in perfect order. The only thing missing? A recipe. You’re not stranded; the freedom might feel exhilarating. A dash of this, a pinch of that, tasting and adjusting as you go. But more often than not, the result is underwhelming. No matter how high-quality the ingredients or cutting-edge the equipment, the magic of a truly exceptional meal often eludes you.
Now, imagine stepping into the same kitchen, this time armed with a recipe crafted by a world-class chef. Step by step, it guides you through the process, showing you how to build flavors and techniques you might never have considered. Suddenly, even a competent cook can produce something remarkable—not Michelin-star quality, perhaps, but far better than you’d achieve alone.
The difference between these two scenarios is not the tools or the ingredients; it’s the structure. A recipe is a map to a higher peak, not just an invitation to explore.
This distinction mirrors the difference between open-ended AI tools like ChatGPT and those powered by structured analytic frameworks. ChatGPT is the well-stocked kitchen without a recipe: flexible, powerful, and capable of turning out useful results. But without structure, it often leads to the digital equivalent of a decent but uninspired meal. It provides options, but without clear direction, those options can blend into the predictable.
Enter structured frameworks, the recipes of thought. These frameworks turn intuition into explicit steps, guiding your process just as a recipe transforms an average cook into someone capable of producing something much greater. A good recipe democratizes expertise, enabling anyone to create something of value. Similarly, structured frameworks empower policymakers, business leaders, and everyday problem-solvers to tackle complexity with tested, repeatable methods.
But here’s the catch: being a great chef and being a great recipe writer are not the same thing. A masterful chef may intuitively know what to do in the kitchen but struggle to translate that skill into instructions others can follow. Similarly, a brilliant thinker may excel at solving problems but falter when asked to codify their approach into a framework others can use. The act of turning tacit expertise into explicit instructions unlocks broader accessibility.
This is where the combination of structured frameworks and generative AI becomes transformative. Some people have the metacognitive skill to generate these repeatable reasoning recipes. Generative AI can then take these frameworks and apply them at speed and scale, allowing many to think like a few. Frameworks ensure that even those without expertise can follow steps, navigate complexity, and reach well-reasoned conclusions.
Take the textbook factory created by Sam Schillace at Microsoft. A user inputs a single sentence about the subject and grade level of the textbook they need. Sam’s program orchestrates 600 API calls to produce a complete textbook, adhering to a recipe for textbook creation: a table of contents, chapters, lesson plans, questions, and answers. The specific ingredients differ depending on whether it’s a math textbook or a history one, but the recipe remains the same. It’s a recipe for AI.
At Dragonfly Thinking, Anthea and Nick are currently working on two case studies applying our repeatable reasoning recipes to the topics of developing an Australian national risk assessment (with John Blackburn) and exploring ways to reduce tribal fighting in Papua New Guinea (with Miranda Forsyth). In both cases, our recipes for thinking enable a complex systems analysis of these very different problems. By breaking down these challenges into structured, repeatable steps, we’re helping to uncover drivers, connections, interventions and feedback loops.
Or take the work of Dimitri at Google Labs who is interested in recipes for thought that are repeatable and therefore scalable. He is developing Breadboard AI as a no code tool to allow people to build these repeatable reasoning recipes both easily and quickly. Many people who are not software engineers have what Dimitri calls "wisdom" that they can impart into LLMs to guide their thinking. Coming up with the recipe as one thing. Coming up with a way to facilitate creating, collecting and distributing thinking recipes is another. We need both for the future thinking.
Just like in the kitchen, the magic isn’t always in the ingredients or the equipment—it’s in knowing how to combine them. A great recipe transforms ordinary ingredients into something memorable. And in the same way, structured thinking frameworks, powered by generative AI, have the potential to elevate our collective thinking—one well-written recipe at a time.