📰 Top News
Anthropic accidentally leaked 512,000 lines of Claude Code's TypeScript source in a routine npm update. Someone rewrote the whole thing in Python and it hit 50K GitHub stars in two hours, creator Boris Cherny's fix for making sure it never happens again is the most ironic part of the whole story 🔓
OpenAI closed a $122B round at an $852B valuation, the largest fundraise in venture history. Their ads pilot hit $100M in annual revenue in under six weeks, and Codex is up 5x in three months 💰
Oracle eliminated up to 30,000 roles via a 6am email this week, no warning, immediate system access cut. The savings go straight into AI infrastructure 📉
🛠️ Product News
Claude Code's CLI can now build, run, and fix your app from one terminal prompt. No switching windows to verify what you built. Computer use is live in research preview for Pro and Max on macOS 🖥️
Zai's GLM-5V-Turbo takes a screenshot of any UI and gives you back working code. It scored 94.8 on the Design2Code benchmark feed it a Figma frame or a web screenshot and it builds the app 🎨
Mac users can now plug in NVIDIA and AMD cards for local AI training. Apple officially approved tinygrad's eGPU driver this week, the hardware excuse for not training locally just got a lot harder to make 🖥️
🔍 Dig Deeper
🔐 The axios supply chain attack this week was designed to be invisible. It installed, ran, and deleted all evidence of itself before you could catch it. Anyone who ran
npm installduring a 3-hour window on March 31 may be affected and who pulled it off is the part that changes everything. Check your lockfile forplain-crypto-js. Here’s karpathy’s take on it:
🔍 The agent stack is maturing and "vibing" isn't going to cut it anymore
Most developers building agents are flying blind. When a multi-step task fails, there's no record of what the agent did, which tool misfired, or where the reasoning broke down.
LangChain published a good framework this week on exactly this. A trace gives you the full behavioral record of what an agent actually did. From there you enrich with evals and human feedback, turn recurring failures into test cases, validate fixes before shipping, and repeat. Reliable agents are built through trace-centered iteration, not one-off debugging.
The number that stuck with us: Claude Code equipped with LangSmith tracing skills jumped from 17% to 92% on their eval set. Check it out here
Ship to Prod Hackathon - SF
Build real AI agents with AWS, Wundergraph & more.
🏆 $40k+ in Prizes!
🗓️ Fri, 24 April
9:30 AM – 7:30 PM PT
Amazon Nova Hackathon - SF
See what devs built on Amazon Nova + hear from Amazon AGI Labs & IBM on agentic AI.
🗓️ Wed, 9 April . 5PM – 8PM PT
Clash of Prompts - SF
Head-to-head AI prompt battles where the best prompt writes the most secure code. Just your prompts, scored live.
🗓️ Thurs, 7 May . 9:30 AM PT
Interested in sponsoring an upcoming event or hosting your own?
reach out here
Overmind — Founding Engineer (London)
Build core ML and security systems from the ground up, including real-time inference pipelines and anomaly detection.
→ ApplyTiger Data — Senior Backend Engineer (Remote)
Build Golang-based services that make data ingestion and export easier for developers and enterprises.
→ ApplyAirbyte — Software Engineer (SF)
Infrastructure and reliability engineering on the Data Replication team, setting standards and driving down incidents.
→ Apply
Promote Your Launch or Hiring Call 🚀
Promote a product launch, event, or hiring call to our community.
Become a Venue Partner 🏢
Host a tokens& event at your office. Get your brand in front of builders, VCs, and sponsors at zero cost.