Ad
Skip to content

Matthias Bastian

Matthias is the co-founder and publisher of THE DECODER, exploring how AI is fundamentally changing the relationship between humans and computers.

OpenAI's chief scientist trusts AI with experiments but says it's not at the level to design complex systems

OpenAI Chief Scientist Jakub Pachocki used to write every line of code by hand. Now AI handles experiments that once took him a week, but he’s not ready to let it run the show.

Read full article about: Nvidia CEO Jensen Huang says he'd be "deeply alarmed" if a $500K developer spent less than $250K on AI tokens

Nvidia CEO Jensen Huang believes that if a developer earns $500,000 a year, their token budget should be at least half that amount. On the All-In podcast at Nvidia's GTC conference, Huang laid out a "thought experiment:" If a developer or AI researcher earned $500,000 a year and only used $5,000 in tokens by year's end, he would "go ape something else." If their token budget wasn't at least $250,000, he'd be "deeply alarmed."

To Huang, it's no different "than one of our chip designers who says, guess what, I'm just going to use paper and pencil. I don't think I'm going to need any CAD tools." The statement has at least as much meme potential as Huang's legendary "The more you buy, the more you save" line from GTC 2018.

On the AI industry's revenue potential, Huang says Anthropic CEO Dario Amodei is "very conservative" with his forecast of hundreds of billions in AI usage revenue by 2027/28 and a trillion dollars by 2030. His reasoning: every enterprise software company will eventually act as a "value-added reseller" of tokens from Anthropic or OpenAI, dramatically expanding the market.

Read full article about: OpenAI acquires Astral to bring Python's most popular dev tools into its Codex AI coding platform

OpenAI is acquiring Astral, the company behind the widely used Python tools Ruff, uv, and ty. Astral founder Charlie Marsh announced that his team is joining OpenAI's Codex team, the company's platform for agentic AI coding. According to Marsh, Astral's tools are downloaded hundreds of millions of times each month and have become a core part of modern Python development. Marsh says integrating them with Codex gives both projects the most room to grow. Astral was backed by Accel and Andreessen Horowitz, among others.

Our goal with Codex is to move beyond AI that simply generates code and toward systems that can participate in the entire development workflow—helping plan changes, modify codebases, run tools, verify results, and maintain software over time. Astral’s developer tools sit directly in that workflow.

OpenAI

OpenAI says it will keep the tools open source after the acquisition closes. Astral's Douglas Creager wrote on Hacker News that the tools are under a permissive license, so in a worst-case scenario, the community could fork the software and continue developing it independently.

No one can guarantee how motives, incentives, and decisions might change years down the line. But that's why we bake optionality into it with the tools being permissively licensed.

Douglas Creager
Read full article about: Anthropic turns Claude Code into an always-on AI agent with new channels feature

Anthropic's Claude Code now supports "channels," letting messages, notifications, and webhooks flow directly into a running session. Claude can respond to events even when the user isn't at the terminal, whether that's CI results, chat messages, or monitoring alerts.

Channels run through MCP servers and support two-way communication: Claude reads an incoming message and responds through the same channel. The research preview supports Telegram and Discord, and developers can build their own custom channels. The feature moves Anthropic's tooling closer to the AI agent hype around OpenClaw.

The feature requires Claude Code version 2.1.80 or later and a claude.ai login; API keys aren't supported. Teams and Enterprise organizations need to explicitly enable channels. Full details are in the official documentation.

Read full article about: Google AI Studio now lets you vibe code real-time multiplayer games

Google has launched a new vibe coding feature in Google AI Studio that lets non-programmers and programmers alike turn ideas into working apps using natural language. Users describe what they want, and Gemini 3.1 Pro handles the technical implementation. Apps are built directly in the browser and can handle things like payments, data storage, or messaging. Google says even multiplayer applications like real-time games are possible.

A new "Antigravity Agent" automatically detects when an app needs a database or login system and sets both up through Firebase. Third-party services like payment providers or Google Maps can be connected using API keys. When needed, the agent also installs web tools like Framer Motion or Shadcn on its own. In addition to React and Angular, the platform now supports Next.js as well.