the end of Claude Code
TL;DR
Anthropic accidentally leaked Claude Code, and the backlash created its own uncontrollable clone — After an April Fool’s update exposed the source, Anthropic sent broad DMCA notices, which helped trigger “claw code,” a clean-room reimplementation that Wes says hit 50,000 GitHub stars in 2 hours and later passed 117,000.
The big story isn’t the copied app — it’s that a complex agent harness was rebuilt in hours — Sigurd Jin, described as having used 25 billion Claude Code tokens, rewrote Claude Code in Python in about 2 hours and says the later Rust rewrite took a day, which Wes frames as a major signal for software development.
Clean-room development just got compressed from a lawyer-heavy process into an AI workflow — Wes explains that copyright protects code, not functionality, using Photopea vs. Photoshop as the analogy, and argues AI can now play both the “dirty team” and “clean team” that used to require separate humans.
Jin’s real claim is that the files don’t matter; the agent system does — The workflow used Oh My Codex on top of OpenAI’s open-source Codex plus a background coordinator called ClawWhip, letting a human drop instructions into Discord while agents split tasks, code, test, debate, fix, and push changes.
This shifts the scarce skill from typing code to architecture and judgment — Wes highlights Jin’s argument that as agents get stronger, the valuable skills become knowing what to build, decomposing work, coordinating multiple agents, and maintaining a clear mental model of the system.
Wes sees this as evidence of a brief era where one person can have outsized leverage — He argues that between AGI and ASI, a single individual at a computer may have more world-shaping agency than at any prior point in history, and presents claw code and OpenClaw as early signs of that shift.
The Breakdown
The leak that turned an April Fool’s gag into AI chaos
Wes opens with the kind of story that sounds made up: Anthropic updated Claude Code, slipped in a Tamagotchi-style April Fool’s feature, and accidentally leaked the whole source code. Within 48 hours, the code had been copied, cloned, and forked everywhere, and Anthropic responded with a “scorched earth” DMCA campaign that set the tone for everything that followed.
Enter Sigurd Jin and the birth of “claw code”
Wes then introduces Sigurd Jin as the perfect chaos agent for this moment — the guy profiled in the Wall Street Journal who reportedly burned through 25 billion Claude Code tokens. After seeing Anthropic’s takedowns, Jin decided to rebuild Claude Code from scratch, and the result — “claw code” — became, in Wes’s telling, the fastest-growing GitHub repo ever, blasting past 50,000 stars in just 2 hours.
Why a legal clone can exist: the clean-room explanation
To make sense of the legality, Wes uses Photopea as the analogy: Photoshop’s code is protected, but Photoshop-like functionality is not. That’s the core of clean-room development — historically a painstaking process with separate “dirty” and “clean” teams — and Wes’ key point is that AI has collapsed that whole structure into something one person can drive at absurd speed.
The part that should scare or excite developers
Wes lingers on what this means emotionally: Anthropic’s elite engineers spent serious time building Claude Code as a valuable proprietary asset, and that advantage got functionally reproduced in hours. He quotes Jin’s line that for some people this feels like a superpower, and for others it looks like “a pink slip,” which captures the mood of the whole episode.
Anthropic’s DMCA overreach and the irony spiral
The takedowns weren’t just aggressive — Wes says they also swept up legitimate repos, including forks of Anthropic’s own open-source projects, which would make some of the notices improper. Anthropic later retracted part of the request and asked GitHub to restore non-infringing repos, but by then the damage was done: the attempt to lock things down had helped create a version they couldn’t legally touch.
The weirdest reveal: agents working through Discord while the human sleeps
The most important section, in Wes’s view, is Jin’s argument that people are focusing on the wrong thing if they stare at the generated Python files. The real breakthrough is the system: a human sends a few sentences in Discord, then agents using Oh My Codex, ClawWhip, and coordination logic break work into tasks, write code, test, argue, fix failures, and push changes — no IDE heroics, just “Discord, a chat.”
From coding skill to system design skill
That leads to Jin’s deeper thesis, which Wes clearly buys: if agents can build fast, then typing speed matters less than architectural clarity, task decomposition, and knowing how pieces should fit together. He even points to a circulating idea that the surviving tech roles cluster around vibe coders, security/infrastructure, people-facing adults in the room, and functions like legal or finance — notably, not traditional code-writing as the center of value.
Wes’ bigger bet: a one-person leverage spike before ASI
Wes ends by zooming all the way out, arguing that there may be a short historical window between AGI and ASI where one person with AI tools can have unprecedented impact. He frames claw code not as a quirky GitHub drama but as evidence that this leverage spike may already be starting, then leaves the audience with the question he thinks matters most: if building gets cheap, what will you build?