The Discovery Stack

For most of the internet era, discovery was a visibility problem.

The challenge was getting found.

Ranking higher.

Generating clicks.

Capturing attention.

If people could see you, you had the opportunity to compete.

Artificial intelligence changes the economics of discovery.

Increasingly, users don’t browse dozens of options.

They ask questions.

Systems interpret intent.

Information is synthesised.

Answers are presented.

Recommendations are made.

The user experiences resolution.

Which raises a different question:

Why do some answers consistently emerge while others disappear?

The answer may lie in understanding the architecture beneath them.

We call this:

The Discovery Stack.

What Is The Discovery Stack?

The Discovery Stack is a way of understanding how AI-mediated discovery works.

It suggests that what users experience as a recommendation or answer is not produced by a single mechanism.

Instead, it emerges from the interaction of multiple layers operating at different speeds.

The user sees the outcome.

The stack explains how that outcome came to be.

The Four Layers

The Index

The index determines whether you can be found.

It includes:

→ websites

→ articles

→ reviews

→ directories

→ databases

→ structured data

The question it answers is:

Can they find you?

If you don’t exist in the index, you rarely enter consideration.

The Context Window

The context window determines what fresh evidence is available at the moment of synthesis.

It incorporates:

→ retrieved sources

→ recent information

→ updated descriptions

→ current proof

The question it answers is:

Can they understand you today?

The context window moves quickly.

It adapts.

It updates.

It reflects the latest version of reality.

The Weights

The weights contain the model’s accumulated understanding of the world.

Its assumptions.

Its inductive biases.

Its priors.

The question they answer is:

Do they already understand you?

The weights move slowly.

They inherit the accumulated sediment of the corpus.

They remember.

Resolution

Resolution is what the user experiences.

The recommendation.

The answer.

The explanation.

The summary.

Resolution is not a separate layer competing with the others.

It is the observable outcome produced by their interaction.

The user never sees the stack.

They see the conclusion.

The Discovery Equation

Resolution emerges from the interaction between:

Index + Context Window + Weights

The same answer can look stable while being produced by very different underlying dynamics.

A recommendation may emerge because:

→ the index is strong

→ retrieval is current

→ the prior is weak

Or because:

→ the prior is dominant

→ fresh evidence simply confirms it

Understanding which layer is doing the work changes the strategy entirely.

The Human Loop

The mechanism is not mystical.

Models don’t simply decide what matters.

They inherit what people wrote about what happened.

Resolution shapes human action.

Human action shapes the corpus.

The corpus shapes future weights.

The internet becomes memory.

The models inherit it later.

What Survives Synthesis?

Increasingly, users don’t consume your entire corpus.

They consume the model’s understanding of it.

Everything becomes input:

→ your website

→ reviews

→ customer stories

→ press coverage

→ structured data

→ recommendations

→ interviews

→ conversations

The system compresses all of this into an answer.

Which means the important question becomes:

What survives synthesis?

Coherence matters.

But coherence alone isn’t enough.

Coherence without specificity becomes invisible.

Specificity without coherence becomes noise.

The strongest priors are both distinctive and sharp.

Say something specific.

Say it everywhere.

The Three Prior Games

Every organisation eventually finds itself playing one of three games.

Build

You don’t yet have a prior.

Your challenge is becoming distinctive enough to matter.

Defend

You have the right prior.

Your challenge is preserving coherence and preventing drift.

Overwrite

You have the wrong prior.

The system understands you.

It simply understands an older version of you.

Your challenge is changing what the world remembers.

The Two Tests

The first question is:

What survives synthesis?

Ask multiple AI systems:

What is this company?

What is it best at?

Who is it for?

Compare the answers.

Look for:

→ hedging

→ contradictions

→ generic descriptions

→ differences between systems

The second question is:

Where do we exist?

Run the same exercise with retrieval enabled.

Then without retrieval.

The difference is revealing.

If the system only understands you when it searches:

You exist in the index.

If retrieval explains you accurately:

You exist in the context window.

If the model already understands you before retrieval begins:

You have a prior.

One caveat matters.

A confident answer is not necessarily a correct one.

Confident and right suggests a prior.

Confident and wrong may suggest:

→ a stale prior

→ a disambiguation problem

The remedies differ.

Both require clarity.

The Bridge Strategy

Different parts of the stack operate on different clocks.

The context window moves quickly.

The weights move slowly.

There isn’t one prior.

There are many.

Different model families.

Different release cycles.

Different training vintages.

You don’t have a prior.

You have a portfolio of priors.

And that is a gift.

Because it makes progress measurable.

Newer models become leading indicators.

Older models become lagging indicators.

You can literally watch understanding propagate through the ecosystem.

While the weights catch up:

Compete to win retrieval.

Use fresh evidence to bridge the gap between who you were and who you’ve become.

Fresh evidence becomes accumulated evidence.

Accumulated evidence becomes corpus.

Corpus becomes weights.

The bridge stops being scaffolding.

It becomes infrastructure.

Why This Matters

The search era asked:

How do we get found?

The Discovery Stack asks:

How are we understood?

Visibility still matters.

Retrieval still matters.

Priors still matter.

But none of them exist in isolation.

They form a stack.

And understanding that stack changes the game.

Because the organisations that thrive in AI-mediated discovery won’t simply be the easiest to find.

They’ll be the easiest to understand.

And ultimately…

the easiest to resolve toward.

Continue Reading

The Discovery Stack is explored in more depth through the following Resolution essays:

The Experiment

The Discovery Stack is now subject to the same forces it describes.

It exists in articles.

Conversations.

Examples.

Critiques.

Responses.

Some readers will encounter fragments.

Others will encounter the whole.

Models may eventually inherit some version of it.

The question is no longer whether it was written.

The question is what survives.

Because the internet becomes memory.

Models inherit it later.