Eating Soup with a Fork

Adam Sturdee
May 23
5 min read

On research, AI, and the coming premium on taste

In her Day 2 keynote at the BERA TEAN conference in Sheffield this week, Professor Eline Vanassche shared a comment that someone had left under one of her LinkedIn posts. Seven words:

"Reality is soup. Research is a fork."

I haven't stopped thinking about it.

It's the kind of throwaway line that quietly rearranges a small filing cabinet in your head. In seven words it captures what I've spent years of school leadership trying to articulate to colleagues, to inspectors, and most often to myself.

The line has a much older sibling. In 1995, Donald McIntyre — writing in the British Journal of Educational Studies in defence of his Oxford "practical theorising" approach to teacher education — argued that the knowledge teachers actually work with is local, temporary, and partial (p373). It was a quiet, devastating sentence then, and it's a quiet, devastating sentence now. Whatever we learn about classrooms is bound to a place, a moment, and a slice. Move it three miles, three years, or three children sideways, and it begins to leak.

Both phrases name something we routinely forget in school leadership: the gap between the knowledge we generate and the reality in which we have to act. McIntyre's point, in essence, was that good teachers don't import theory; they theorise practically, in context, on the move. Thirty years on, the case has only got stronger.

The fork problem in school improvement

Walk into any senior leadership meeting in England and you will hear the EEF cited, the Sutton Trust invoked, a Sherrington tweet quoted, a Rosenshine principle referenced. None of this is wrong. The fork is a useful instrument. You cannot eat soup without something, and a fork is, at minimum, something.

But the problem with leading a school by fork is that you end up trying to reconstruct the soup from what you've picked up — and the broth, which is most of the meal, runs back into the bowl.

The retrieval practice study was conducted with these students, this age, this subject, this teacher, this term, this country. The metacognition intervention was effective at a particular dosage, with a particular fidelity, in a particular climate. The coaching evidence base is real, but the schools studied had their own cultures, the samples were self-selected, the contexts were not yours. Local. Temporary. Partial.

This isn't an argument against research. It's an argument against the genre of school improvement that mistakes findings for instructions — that takes the fork home and tries to serve soup with it.

The best senior leaders I know hold research the way good cooks hold a recipe: with respect, and with the unspoken understanding that they will deviate from it the moment the kitchen tells them to.

AI is also a fork

Here is where it gets interesting for those of us working at the AI–education seam.

Every claim being made about generative AI in classrooms right now is forked in two ways. First, the underlying evidence: studies of ChatGPT use, of marking accuracy, of pupil engagement, are themselves local, temporary, partial — often run on undergraduates, in artificial conditions, with prompts a tired teacher would never write. Second, the models themselves are forks of reality: training data is selective, benchmarks are narrow, capabilities shift weekly. We are forking a fork.

This doesn't mean AI has nothing to offer schools. I've staked a meaningful chunk of my professional life on the conviction that it does. But it means the implementation question is not "what does the research say?" It is "what does my school's soup actually taste like, and does this tool make it better?"

The schools getting AI right are not the ones with the strongest evidence base. They are the ones whose leaders have a developed palate — who can taste, in a Year 9 essay or a coaching conversation or a marking pile at 9pm on a Tuesday, what is actually going on. From that palate they make small, calibrated additions. Sometimes the addition is a Claude project. Sometimes it's just deciding not to add anything yet.

Senior leadership lives in the soup

Researchers fork. Inspectors fork. Consultants fork. Senior leaders eat.

The job is not to read the soup; it is to live in it. You feel it on your face when you walk through reception in October. You know it in the corridor outside the Year 11 toilets. You hear it in the staffroom on a Friday in February. None of that is in a paper. All of it is the actual material on which your decisions act.

This is what McIntyre was getting at. The practical theoriser doesn't choose between theory and experience; they hold both in tension, in context, in real time. They use theory the way a cook uses salt — to season what's actually in the pan, not as a substitute for the dish.

This is why the leadership skill that matters most is not synthesis of evidence — that's the analyst's job — but judgement under conditions of irreducible partiality. Knowing which forkful to trust this week. Knowing when the recipe is wrong for this kitchen. Knowing when to put the recipe down altogether.

The coming premium on taste

Which brings me to the part of this argument I think is underappreciated.

In a world where AI can produce competent, plausible, well-structured outputs at near-zero marginal cost — schemes of work, parent letters, lesson plans, even reasonable research summaries — the scarce thing is not production. It is discrimination. It is taste.

Taste is the human capacity to know, often without being able to explain why, that this is right for here, and that isn't. It is the soup-side faculty. It doesn't run on findings; it runs on time spent, attention paid, mistakes survived. A head of department who has marked ten thousand essays has a kind of knowledge no benchmark captures. A pastoral lead who has sat with hundreds of crying Year 10s has a kind of knowledge no evidence base will ever overtake.

Authenticity is the public expression of that taste. It is the reason a parent reads one head's newsletter and trusts it, and reads another's and skims. It is the reason some teachers' lessons feel like theirs and others feel like a download. AI raises the floor and, paradoxically, raises the value of everything that can't be produced by it.

For school leaders, this has a practical consequence. The next decade will not reward the leader who is the best at reading research. It will reward the leader who has the most developed palate — who has tasted enough soup, in enough kitchens, to know what theirs needs. Research will still matter; it just won't be the main course. It will be one ingredient among many, including ones we don't yet have a word for.

A small confession at the end

I write this as someone who reads more research than is probably good for him, runs an EdTech company built on AI, and spent this week in lessons watching teachers do things no paper will ever describe. The contradiction is the point. We need the forks. We also need to stop pretending they are the meal.

Reality is soup. Research is a fork. AI is a faster fork.

Taste is the cook.

Adam Sturdee is a senior leader and co-founder of Starlight, the UK’s teacher-first AI-powered transcript-based coaching platform for educators.

His work sits at the intersection of dialogic practice, instructional leadership and responsible AI strategy for schools and trusts.

He recently presented his research on AI-supported coaching at the BERA TEAN Conference 2026: https://www.bera.ac.uk/conference/bera-tean-conference-2026

Eating Soup with a Fork

Recent Posts

Comments