📰 News
-
Claude Code + Fable 5 ported a 2003 PC game to iOS in hours — Google DeepMind developer Ammaar Reshi used Claude Code with Fable 5 to port Command & Conquer: Generals Zero Hour natively to iPhone and iPad (ARM64, no emulator). First build took 40 minutes; two days of debugging burned through his entire Claude Max quota. DirectX 8 → Apple Metal translation included. Full source is open source. When asked why he used a competitor’s tool, he said: “You can love the AI space and respect the competition.” → Read more
-
Alibaba reportedly bans employees from using Claude Code — Effective July 10. With Anthropic already prohibiting Chinese companies from accessing Claude, Alibaba classified Claude Code as “high-risk software” and told staff to switch to its in-house Qoder tool instead. → Read more
-
Zuckerberg tells staff AI agents haven’t moved as fast as hoped — At an internal town hall, Meta’s CEO admitted the AI agent rollout hasn’t “come to fruition yet.” Meta laid off 8,000 employees earlier this year, reassigned 7,000 more to AI groups — yet the payoff is still 3–6 months away, he says. The company plans to spend $145 billion on AI infrastructure this year alone. → Read more
-
Mistral CEO: closed AI models give labs a front-row seat to your business — Arthur Mensch posted on LinkedIn that proprietary AI vendors accumulate enough customer data to understand — and potentially replicate — their customers’ business processes. He has a point, though Mistral’s own open-source model business makes this argument self-serving. Palantir’s Alex Karp made a similar case recently. → Read more
-
Hollywood wants Seedance banned — and quietly keeps using it — ByteDance’s AI video tool Seedance is the subject of a cease-and-desist from the Motion Picture Association, yet studios are reportedly tolerating it on a “don’t ask, don’t tell” basis. Creatives who use it call it the best video tool on the market right now. → Read more
-
AI search agents fail at asking, not at searching — A Tencent/Tsinghua benchmark (DiscoBench) found that even the best models top out below 50% accuracy on multi-hop research tasks. The bottleneck isn’t the search itself — it’s the model’s inability to ask users for clarification when a query is ambiguous. Searching more actually hurts: models that search repeatedly without clarifying perform worse than those that just guess. → Read more
-
Hugging Face + Cerebras demo real-time voice AI with Gemma 4 — A fully open, modular speech-to-speech pipeline: Nvidia Parakeet (STT) → Gemma 4 31B on Cerebras → Qwen3TTS. Cerebras’s fast inference makes the conversation feel natural — the same pipeline already runs on 9,000+ Reachy Mini robots in the wild. → Read more
🔥 GitHub Trending
-
openai/codex-plugin-cc — A plugin that brings OpenAI Codex into Claude Code for code reviews and task delegation. Two competing AI coding tools, one workflow — surprisingly practical.
-
alibaba/page-agent — In-page JavaScript GUI agent that lets you control any web interface with natural language. No browser extension or headless browser needed — just drop in a
<script>tag. Includes an MCP server for external agent control. -
ChromeDevTools/chrome-devtools-mcp — Official Google MCP server that gives AI coding agents (Claude, Cursor, Copilot) full access to Chrome DevTools for debugging, network inspection, screenshots, and performance tracing.
-
asgeirtj/system_prompts_leaks — Regularly updated collection of leaked system prompts for Claude Fable 5, Opus 4.8, Claude Code, GPT-5.5, Gemini, and more. Featured in The Washington Post.
-
JuliusBrussee/caveman — A Claude Code skill that makes your AI agent talk like a caveman — 65% fewer output tokens, 100% technical accuracy. Strips the filler, keeps the code.
▶️ YouTube
- They Said This Will Never Run In Real Time — Two Minute Papers | The channel’s classic “impossible → possible” format, covering a paper about something once deemed too slow for real-time that now runs in real-time.
💬 Community
-
sqlite-utils 4.0rc2, mostly written by Claude Fable for ~$149 — Simon Willison handed the final pre-release review to Fable, which caught 5 release-blocking bugs including a data-loss issue in
delete_where(). 37 prompts, 34 commits, 30 files changed. Willison ran it from his phone during the 4th of July parade. Total cost: $149.25. -
Better Models, Worse Tools — A developer found that Claude Opus 4.8 and Sonnet 5 are worse at third-party edit tools than older models. Likely because newer models are RL-trained specifically on Claude Code’s own edit tool schema. Third-party coding harnesses may need to offer multiple edit tool schemas depending on which model the user has selected.
-
Open Source AI Gap Map — A structured index of 421 open-source AI products: 266 software tools, 85 models, 50 datasets, 20 hardware projects across 228 organizations. MIT-licensed YAML data (1,184 files) released on GitHub.