Skip to main content

The Secret That Wasn’t - Or: What Happens When the Question Finally Appears

When the internal architecture behind Claude Code briefly appeared online earlier this year, the internet reacted in the way the internet always does.

Someone rebuilt it.

Reports suggested a working clone appeared within roughly two hours. Not copied line by line, but reconstructed once the underlying pattern became visible. What emerged became known as Claw Code, a recreation inspired by the development agent built by Anthropic.

The speed of the reconstruction was startling, but perhaps not surprising.

Because the most important part of the story is not the leak.

It is the moment the question finally appeared.


When the Mystery Disappears

Before the leak, Claude Code existed largely as a black box. Observers knew what it did. They could see its capabilities. They could speculate about its architecture.

But speculation is not the same as understanding.

Then the curtain lifted.

Developers saw something that looked far less mysterious than expected: an orchestration system where a language model coordinated tools, inspected code, executed commands and evaluated outcomes in a loop.

Once the structure was visible, replication followed quickly.

This pattern repeats throughout the history of technology. Once architecture becomes legible, the mystery dissolves.

The “secret sauce” turns out to be a recipe.


The Architecture of Intelligence

What the leak revealed was not a single miraculous algorithm. It was a design.

Planning loops.
Tool invocation.
Memory management.
Iterative evaluation.

Anyone who works in programme delivery or organisational change would recognise the structure immediately. It looks suspiciously like a workflow. Tasks are broken down, actions are taken, feedback is gathered and the process repeats.

In other words, the system resembles a small organisation encoded in software.

This is the deeper shift happening in artificial intelligence. The systems that appear most intelligent are often the ones that are best organised.

Intelligence emerges from coordination.


The Fiction of Control

In a previous essay I wrote about what I called the fiction of control: the tendency of organisations to construct narratives of mastery over systems that are far more fragile than they appear.

The Claude Code episode fits neatly into that pattern.

Technology companies understandably protect their intellectual property. They guard repositories. They manage access. They treat architecture as strategic advantage.

Yet once the design of a system becomes visible, its defensibility changes dramatically.

What had appeared proprietary becomes reproducible.

The illusion was not the technology. The illusion was the exclusivity.


A Familiar Scene

There is a moment early in Jaws when the town’s leadership insists the beaches must remain open. The evidence of danger is mounting, but the economic narrative is powerful. Officials reassure themselves that the situation is manageable.

The shark, as it turns out, has its own views.

Organisations often behave in a similar way with technology. The story of control becomes easier to maintain than the reality.

When Claude Code’s architecture briefly surfaced, the industry experienced something rare: a moment where the curtain moved aside.

What developers saw was not magic. It was structure.


Learning at Internet Speed

The rapid emergence of Claw Code also demonstrated something fascinating about how knowledge now moves.

Developers analysed the architecture.
They tested variations.
They rebuilt components.
They shared improvements.

The global software community behaved like a massive distributed learning system.

This is what Peter Senge meant by a learning organisation, although he probably did not imagine one consisting of thousands of engineers connected through GitHub and messaging platforms.

Once architecture enters that system, adaptation accelerates dramatically.

Which brings us back to the question.


The Question That Was Missing

In another piece I asked whether it might be better to have a question without an answer than an answer without a question.

The Claude Code story illustrates the difference beautifully.

For months the industry had answers.

Claude Code works.
Claude Code is powerful.
Claude Code is proprietary.

But the key question had not yet appeared.

How is it actually organised?

Once that question emerged, the answers multiplied quickly.

The architecture became legible. The mystery disappeared. Engineers rebuilt the system.

Two hours later, the industry understood something it had only previously imagined.


Intelligence as Organisation

The deeper lesson may be philosophical.

Artificial intelligence is often framed as a competition over models. Which system is smarter? Which algorithm is more capable?

Yet the Claude Code episode suggests something different.

The most powerful systems may not simply be those with better models. They may be the ones that organise intelligence more effectively.

Planning.
Coordination.
Iteration.
Feedback.

Those are not only technical concepts. They are also organisational ones.

Which means the disciplines of governance, leadership and change management may have more to say about AI than we sometimes assume.


Once the Recipe Escapes

When the architecture behind Claude Code became visible, developers realised they already knew how to build most of the pieces.

They simply had not seen the recipe before.

Once they had, the internet did what the internet always does.

It started cooking.

The organisations that thrive in this environment may not be the ones that hide their ideas most carefully.

They may be the ones that learn fastest once the recipe escapes.

Comments

Popular posts from this blog

✨ Act III – Frills and Fills ✨ AI Exceptionalism: Myths, Motives, and Mundane Realities.

We’ve reached the final act of AI Exceptionalism: Myths, Motives, and Mundane Realities. In Act I, we set the rhythm—how AI was cast as an actor, why neutrality is a myth, and how politics turns every briefing into a bit of theatre. Act II followed the notes—the human cost of hype, the inheritance of bias, and the wobble that came when GPT-5 didn’t quite match its own legend. Now Act III begins. This is the act of frills: the solos, fills, and ornamental runs that dazzle the ear but can drown the bassline. Over the next few scenes, I’ll ask: * Why legislation often protects incumbents more than citizens * Whether the AI bubble bursts or just flattens into everyday routine * What happens when every “exceptional” technology fades quietly into the background Each scene will be short. No neat answers—just nudges worth thinking about. Scene 7 – Why Legislate? Whose Interests? Laws aren’t written in the calm of reflection; they’re drafted in the noise of headlines. AI is no different. Govern...

Best for the Role III = Beyond the Mirror — Leading Diverse Thinking from a Homogenous Top

You can tell a lot from a photo. A leadership team lined up before a neutral backdrop. Navy, grey, black — an orderly composition of confidence and calm. Everyone looks capable. Everyone looks… similar. It’s professional, deliberate, reassuring. But to a truly diverse workforce, it can look like a closed circuit. No light, no air, no movement. That’s the paradox of leadership imagery today: the same photo that reads as competence to some can read as distance to others. And as Roseline asked in a recent thread — a question that stayed with me — W hat does it mean to be visible as a leader? Is visibility about recognition, or connection? About being seen , or being known ? The question matters, because if your leadership team looks homogeneous, the temptation is to manage the optics — to fix the photo. But real inclusion doesn’t happen in the frame. It happens in how you design the lens, the focus, and the depth of field. 1. The Still Image and the Moving System A photo fre...

AI Exceptionalism Act I: Myths, Motives, and Mundane Realities - Act I - Rhythm (the pulse beneath everything)

Act I – Rhythm (the pulse beneath everything) Rhythm is about timing and emphasis. It’s the heartbeat that makes things move. These first essays set that beat: how AI is framed, why it feels urgent, and why the tempo keeps rising. Scene 1 – The Tool That Became an Actor A councillor once asked me, half-serious, “Can AI be sued for lying?” It stuck with me. Nobody ever wondered that about Microsoft Word. No-one wrote legislation for autocorrect or Clippy when it quietly changed definately to definitely. Yet when a chatbot drafts a legal opinion, ministers start talking about treaties and guardrails. Same mechanism, different performance. The difference isn’t technical; it’s theatrical. Word is a screwdriver—predictable, blunt. AI walks on stage, speaks lines, demands applause. Once a tool is cast as an actor instead of a prop, we start assigning it motives: Could it mislead? Could it betray? Might it refuse to open the pod-bay doors? We’ve rehearsed this story for decades. HAL 9000 frig...