Website explains how anthropic claude code works

[Jason Voorhees]

I won't waste time talking about it because if you are truly interested and have technical knowledge you already knew what was going on or have already watched a video of some YouTuber breaking it down but I will share this website that someone recently vibe coded explaining the entire process

Claude Code Unpacked

What actually happens when you type a message into Claude Code? The agent loop, 50+ tools, multi-agent orchestration, and unreleased features, mapped from source.

ccunpacked.dev

In all chatbots is the hidden prompt that the system already has gets combined with your own prompt then gets fed into the LLM and out comes AI slop but in claude the process is a lot more complex and more involved.

There's a full agent loop, 50+ tools for file ops / bash / planning / multi-agent stuff, slash commands, state management, and all kinds of orchestration layers on top of the model so it the answer it spits out seem much better.


TLDR- Claude is not magic. It just uses a bunch of concepts that were known since 1970s and sandwiches then with thousands of lines prompt spaghetti with hard guardrails in place to ensure the LLM doesn't something weird. No superhuman intelligence or something revolutionary happening. Just massive scale, tons of context and serious computing power behind it that makes it feel like magic.

 

[iblamechico]

Hymen

 

[unknownincel]

sorry unc bot interested

 

[iblamechico]

is claude the best for vibecoding currently?

 

[Jason Voorhees]

@Chadeep @Framemaxxx @wunderkind @cash.org @estrogen consumer

 

[Jason Voorhees]

@thekidd8 @Sprinkles @Bryce

 

[wunderkind]

I won't waste time talking about it because if you are truly interested and have technical knowledge you already knew what was going on or have already watched a video of some YouTuber breaking it down but I will share this website that someone recently vibe coded explaining the entire process

Claude Code Unpacked

What actually happens when you type a message into Claude Code? The agent loop, 50+ tools, multi-agent orchestration, and unreleased features, mapped from source.

ccunpacked.dev

In all chatbots is the hidden prompt that the system already has gets combined with your own prompt then gets fed into the LLM and out comes AI slop but in claude the process is a lot more complex and more involved.

There's a full agent loop, 50+ tools for file ops / bash / planning / multi-agent stuff, slash commands, state management, and all kinds of orchestration layers on top of the model so it the answer it spits out seem much better.


TLDR- Claude is not magic. It just uses a bunch of concepts that were known since 1970s and sandwiches then with thousands of lines prompt spaghetti with hard guardrails in place to ensure the LLM doesn't something weird. No superhuman intelligence or something revolutionary happening. Just massive scale, tons of context and serious computing power behind it that makes it feel like magic.

the actual cool shit that dropped this week is turboquant, i gotta go and read that paper man

 

[Jason Voorhees]

@lnceIs @Glorious King

 

[Bars]

Yeah, that's what I was actually thinking, even tho Idrk much about this. Llm is overrated

 

[Glorious King]

I won't waste time talking about it because if you are truly interested and have technical knowledge you already knew what was going on or have already watched a video of some YouTuber breaking it down but I will share this website that someone recently vibe coded explaining the entire process

Claude Code Unpacked

What actually happens when you type a message into Claude Code? The agent loop, 50+ tools, multi-agent orchestration, and unreleased features, mapped from source.

ccunpacked.dev

In all chatbots is the hidden prompt that the system already has gets combined with your own prompt then gets fed into the LLM and out comes AI slop but in claude the process is a lot more complex and more involved.

There's a full agent loop, 50+ tools for file ops / bash / planning / multi-agent stuff, slash commands, state management, and all kinds of orchestration layers on top of the model so it the answer it spits out seem much better.


TLDR- Claude is not magic. It just uses a bunch of concepts that were known since 1970s and sandwiches then with thousands of lines prompt spaghetti with hard guardrails in place to ensure the LLM doesn't something weird. No superhuman intelligence or something revolutionary happening. Just massive scale, tons of context and serious computing power behind it that makes it feel like magic.

i fucking knew it

i always had a feeling that anthropic is just using high crafted prompts bundled into a package with elite tier guard rails


cus no wayy their model could outperform everyone by not upgrading their compute

 

[cash.org]

interesting either way it’s still number 1