I Asked ChatGPT About Claude Code. It Made Up an Entire Theory.

By

So here's what happened.

I was coding with Claude Code and noticed these little status words that pop up while it's thinking: Crafting, Thinking, Roosting, Metamorphosing, Percolating...

Pretty cool. I wanted to know what each one meant — like, do they represent different processing stages? So I asked ChatGPT.

ChatGPT Came In Hot

Instant response. Super detailed. Dripping with authority.

It told me:

These status words map to a fixed state machine with four distinct phases:

Planning / Crafting → Thinking / Reasoning → Roosting / Finalizing → Metamorphosing / Formatting

It's essentially a visualization of the Agent Loop.

It even drew a flowchart:

[Input]
   ↓
Planning / Crafting
   ↓
Thinking / Reasoning / Analyzing
   ↓
Tool Use (Executing / Fetching)
   ↓
Roosting / Finalizing
   ↓
Metamorphosing / Formatting
   ↓
[Output]

And added:

  • "The word pool contains approximately 10-20 tokens"
  • "Weakly correlated with reasoning depth"
  • "An abstraction built by Anthropic at the agent layer"
  • "No public docs, but can be considered a soft-defined stable set"

I read all that and thought: wow, that's way more sophisticated than I expected.

Then I Asked Claude Code Itself

Claude Code's answer was one line:

They're random. Picked from an array. No correlation to processing stages whatsoever. Pure decoration.

What ChatGPT saidReality
Maps to a fixed state machineNo state machine. Random display.
Four ordered phasesCompletely unordered
10-20 token pool with dynamic compositionJust a list, random pick
Weakly correlated with reasoning depthZero correlation
Agent Loop visualizationNothing to do with the agent loop

ChatGPT looked me dead in the eye and, with perfect formatting, clear logic, and professional terminology, fabricated an entire theory that doesn't exist.

And it was good. Good enough that I almost bought it.

A Quick Note on Anthropic's Vibe

Here's the thing — those random loading words are actually kind of beautiful. Roosting (a bird settling in for the night). Percolating (coffee slowly dripping through a filter). Metamorphosing (a butterfly breaking free). Beaming (radiating light).

They turned a boring "Loading..." into a tiny poem.

That's very Anthropic.

They name their models the same way: Opus (a grand symphonic movement), Sonnet (a 14-line poem), Haiku (a 3-line Japanese verse). Not GPT-4o or GPT-4o-mini — engineering serial numbers — but literary forms ranked by depth. The most powerful is a symphony. The lightest is a haiku. In between, a sonnet.

An AI safety company with a poet's soul.

This sensibility bleeds into every corner of the product. The loading animation doesn't say "Processing..." — it says "Roosting..." A bird is coming home. Ideas are brewing. Zero functional purpose, but it makes you smile.

And that's exactly why ChatGPT misread these words as "technical stages" — in OpenAI's worldview, every UI element must serve an engineering purpose. It literally cannot comprehend "just for fun."

Why This Matters

I'm not here to trash ChatGPT. It's a great tool. I use it too.

But this case is textbook — a perfect demo of what AI hallucination actually looks like in the wild:

1. It never says "I don't know"

Claude Code's loading animation is an obscure implementation detail. The correct answer is "not sure, check the source code." But ChatGPT chose to invent an answer instead.

2. The fabrication is extremely convincing

State machines, Agent Loops, token pools, weak correlations — every single concept it used is a real CS term. It just assembled real concepts into a fake narrative.

3. The more polished the structure, the more dangerous it is

If it had just said "probably different stages or something," I would've been skeptical. But it gave me flowcharts, tables, code examples, classification systems — and that perfect structure is exactly what made me drop my guard.

The Coincidence

Today — the same day this happened — Anthropic published a survey of 81,000 Claude users.

The #1 concern?

AI unreliability (26.7%) — hallucinations, fake citations, fabricating plausible-sounding explanations.

I literally became that 26.7% today.

The report also found that lawyers are simultaneously the most benefited and most harmed user group — because AI-generated legal text looks so real that if you can't verify it, you're cooked.

Same deal here. If I hadn't been able to ask Claude Code directly for cross-verification, I might've published a whole blog post called "Deep Dive: Claude Code's State Machine Architecture" — and misled every single reader.

The Takeaway

The most valuable skill in the AI era isn't "using AI" — it's verifying AI.

ChatGPT wasn't trying to deceive me. It was doing exactly what it was trained to do: produce a plausible-sounding response. It doesn't know it's making stuff up — like a student BSing an exam answer they don't know, writing more and more to sound convincing, while the core is completely wrong.

So:

  • Default to skepticism with AI-generated info
  • Cross-verify anything important
  • The more polished the answer, the harder you should push back
  • Best verification method: ask a different AI, or just read the source code

Crafting... Thinking... Roosting... Metamorphosing...

These words mean nothing.

But the theory built around them almost did.

Related: What 81,000 People Want from AI — Anthropic's full survey report

Comments

Post a Comment

Popular posts from this blog

One-Person Software Company: The AI Trinity Method (Part 1 of 3)

By
One-Person Software Company: The AI Trinity Method (Part 1 of 3) Stop Chatting with AI. Start Conducting It. This is Part 1 of a 3-part series on building a production-grade AI development workflow. Part 2: The Shared Brain Protocol→ | Part 3: The Human Conductor → 2 AM on a Wednesday A solo developer has three windows open on his screen. In the left window, he’s debating system architecture with an AI playing the role of a paranoid chief architect — one that challenges every design decision with: “If this has a bug, is it cheaper to fix today, or after a hundred modules depend on it?” In the middle window, a different AI is reviewing code line by line against the architect’s specs, nitpicking like a strict tech lead who’s seen too many production outages. In the right window, a third AI is quietly writing code in a local terminal — running tests, committing to Git, doing exactly what the first two told it to do. All three AIs are Claude. And the person behind the keyboard i...
Read more

Software Company: The AI Trinity Method (Part 2 of 3)

By
The Shared Brain Protocol This is Part 2 of a 3-part series. ← Part 1: Stop Chatting, Start Conducting | Part 3: The Human Conductor → The Recap In Part 1 , we introduced the AI Trinity — splitting AI into three specialized roles (Architect, Tech Lead, Engineer) instead of treating it as a single do-everything assistant. The result: separation of concerns, natural quality gates, and cross-model review that catches bugs no single model would find alone. But we ended on a problem: AI has no memory across conversations. Three brilliant team members who forget everything the moment you close the tab. Today we fix that. The Amnesia Problem Here’s what actually happens without a solution: Monday : You spend two hours with the Architect designing a module. You agree on interface contracts, data formats, error handling strategies. Great session. Tuesday : You open a new chat with the Tech Lead to start implementation guidance. The Tech Lead has never heard of Monday’s decisions. Yo...
Read more

One-Person Software Company: The AI Trinity Method (Part 3 of 3)

By
The Human Conductor This is Part 3 of a 3-part series. ← Part 1: Stop Chatting, Start Conducting | ← Part 2: The Shared BrainProtocol The Story So Far In Part 1 , we split AI into three specialized roles — Architect, Tech Lead, and Engineer — because the AI that writes the code should never review its own code. In Part 2 , we solved AI’s amnesia problem with the Shared Brain Protocol — a structured state document that gives any AI instant full-project awareness. Now for the final piece: you. Because here’s the thing nobody tells you about AI-augmented development: the human’s job doesn’t get easier. It gets different . And arguably more important. You Are Not the Operator. You Are the Commander. In traditional AI usage, the human is the operator — hand-crafting every prompt, manually inspecting every output, doing the cognitive heavy lifting with AI as a fancy autocomplete. In the Trinity method, the human is the commander — making strategic decisions, routing tasks, and ...
Read more