Anthropic’s Claude Code Source Leak Takes Center Stage

#1 📝 Anthropic Leak

By far the biggest AI news today is the leak of Anthropic’s Claude Code: a 500k+ LOC codebase now publicly forked. While not overtly harmful, it’s an embarrassment for Anthropic and a goldmine for coding agent enthusiasts. Coverage highlights curiosities like the full verb list, Capybara/Mythos v8, the '/buddy' April Fools feature, Boris’s WTF counter and the emergent “Claude Codex.” Key technical takeaways:

• Repo-state context (commits, branches)
• Aggressive cache reuse
• Custom grep/glob/LSP integrations
• Under-20 default tools (e.g., AgentTool, BashTool, FileReadTool…)
• File-read deduplication & result sampling
• Structured session memory & subagents

#2 📝 OpenAI News

OpenAI closed its record-breaking $122B fundraise with a soft IPO via private investors and ARK Invest ETFs. The funding announcement highlighted a $4B ARR—4× faster initial growth than Google/Meta—$2B monthly revenue, 900M weekly active users (still below the 1B 2025 target), and $100M ad revenue in six weeks. Notably, Codex reported no new milestone for March.

#3 𝕏

Altryne wrote: If you, like me, just woke up, let me catch you up on the Claude Code Leak (I know nothing, all conjecture):

• Someone inside Anthropic was switched to Adaptive reasoning mode, triggering their Claude Code to switch to Sonnet
• They committed the .map file of Claude Code, effectively leaking the ENTIRE CC source code
• @realsigridjin, tired after running two South Korean hackathons in SF, spotted the leak
• With different rules in Korea, he cloned the repo and went to sleep
• He woke up to 25K stars and his girlfriend (a copyright lawyer) begging him to take it down
• The team then used AI agents to rewrite the code in Python—"surely this is more legal"
• They boarded a plane to South Korea
• One member now finds Python too slow and is rewriting ALL of Claude Code into Rust
• Anthropic can’t easily take it down or sue—Is this fair?

#4 𝕏

Logan Kilpatrick introduced Veo 3.1 Lite, Google’s lightweight on-device AI model offering fast, high-quality language processing at low latency and minimal memory footprint.

Also covered by: @Logan Kilpatrick

#5 𝕏

Google Research launched a new evaluation framework that optimizes the ratio of benchmark items to human raters per item, improving capture of nuanced disagreement in subjective AI tasks and boosting reproducibility.

#6 in

Claire Vo spent 100+ hours setting up OpenClaw—tweaking docs, source code, and configs on four Mac Minis with peers—and distilled her entire process into “The Ultimate 0 > 🦞 Guide” on Lenny Rachitsky’s newsletter.

Also covered by: @Lenny Rachitsky

#7 in

Udi Menkes built an AI agent, Zai, that ingests a daily, optimized brief from high-signal sources into actionable, three-field insights. It ranks updates by impact and applies top priorities like integrating Plastic Labs’ Honcho for episodic memory.

#8 𝕏

Harrison Chase explains how to power a continual agent improvement loop with Langsmith, using trace-centered iteration from LangChain’s “agent improvement loop” guide.

#9 𝕏

bolt.new launched Design System Agents: upload your npm package, @storybookjs, repo, etc., and Bolt builds prototypes using your design system code. Engineering then wires up the backend and ships without discarding prototypes.

#10 𝕏

Jason Zhou launched direct Codex integration in Claude Code, enabling CC code reviews via four simple plugin commands—/plugin marketplace add openai/codex-plugin-cc, /plugin install codex@openai-codex, /reload-plugins, and /codex:setup.

#11 𝕏

Lenny Rachitsky highlights Claire Vo’s definitive “OpenClaw” guide, walking you through first installs, multi-agent setups, plus real-world cost breakdowns and security gotchas.

Also covered by: @Lenny Rachitsky

#12 in

Colin Matthews highlights that LLMs use fixed-size context windows (now up to ~4 million words) and require full chat histories for accurate responses.

#13 in

Anu Jagga Narang built eval rubrics tracking accuracy, hallucinations, and toxicity across every customer conversation. She argues that while evals may be seen as the new PRDs, the real work remains defining the product’s purpose, audience, and success criteria.

#14 📝 OpenAI News

OpenAI raises $122 billion to accelerate the next phase of AI - OpenAI announced it raised $122 billion to accelerate the next phase of AI. The announcement frames the funding as enabling expanded research, infrastructure, safety work, and product development to scale AI capabilities.

#15 in

Dharmesh Shah reports OpenAI’s latest $122 billion funding round values it at $852 billion, with $2 billion in monthly revenue, 900 million weekly active users, $100 million in ad revenue in six weeks, and $3 billion from individual backers.

#16 in

Udi Menkes reveals that leaked Claude source code includes KAIROS, a hidden 24/7 AI agent that self-triggers to fix bugs, update files, send push notifications, deliver new assets, and monitor GitHub PRs.

#17 𝕏

bolt.new launched Design System Agents to turn your repos, npm packages, and docs into agents that build engineering-ready, shippable prototypes (video demo included).

#18 📝 Surge AI Blog

Hemingway-bench Leaderboard: Because Good Writing Isn't a Checklist of Vibes - Hemingway-bench is an AI writing leaderboard that evaluates models on real-world writing tasks judged by master wordsmiths to encourage nuance and impactful prose rather than shallow stylistic signals. The project aims to push AI writing beyond quick 'vibes' toward genuinely high-quality writing.