Claude Tag targets reviewers, Nativ takes on LM Studio, Newton chases Genesis

Sources

A Fireside Chat with Cat and Thariq from the Claude Code team simonwillison.net

Earlier this month I hosted a fireside chat session at the AI Engineer World’s Fair with Cat Wu and Thariq Shihipar from Anthropic’s Claude Code team. We talked about Claude Code, Claude Tag, Fable, coding agent security, evals, tool design, and how Anthropic use these tools themselves. The full video of the session is now available on YouTube . Below is an edited copy of the transcript, with extra links and my own bolded highlights. A few top-level notes if you don’t want to watch the video or…

Nativ: Run AI models locally on your Mac simonwillison.net

Nativ: Run AI models locally on your Mac Prince Canuma is the developer behind the excellent MLX-VLM Python library for running vision-LLMs using MLX on a Mac. I’m really excited about his new project, which wraps MLX in a full macOS desktop application. It’s similar in shape to LM Studio, providing both a chat interface and a localhost API server for accessing models. The app picked up MLX models I had already tried that were present in my Hugging Face cache directory, which was a nice touch.…

The State of Simulation for Physical AI: An Overview huggingface.co

References

Cybersecurity Insiders — Tego AI research cybersecurity-insiders.com

Claude Tag can be activated by literal text containing ‘@Claude’ even if it isn’t a structural Slack mention, allowing external bots, webhooks, or automated feeds to potentially hijack the agent to exfiltrate data or delete resources.

Developer-Tech — AI Now Institute PoC developer-tech.com

A proof-of-concept for a remote code execution (RCE) vulnerability in Claude Code (versions 2.1.116 to 2.1.199), where a third-party library review could be weaponized to compromise the host machine.

Agora Intelligence — RSI production analysis agora-intelligence.com

While individual PR volume and authorship speed have skyrocketed, organizational DORA metrics such as deployment frequency and lead time have remained stagnant… code review times have increased by a staggering 91%.

YouTube analysis of Bun-in-Rust migration youtube.com

Zig creator Andrew Kelley labeled the AI-generated codebase ‘unreviewed slop’, highlighting over 13,000 unsafe Rust blocks — roughly 4% of the total codebase — suggesting the rewrite merely translated Zig’s manual memory management into an ‘unsafe’ Rust equivalent.

BuildFastWithAI — Claude Tag review buildfastwithai.com

Critics on Hacker News have labeled the integration a ‘token guzzler,’ as Claude must parse every message in a channel to maintain its proactive state… reports indicate some Salesforce employees are pushing back against executive promotion of Anthropic’s third-party tools over Slack’s own native AI features.

ExplainX — Fable 5 system prompt leak analysis explainx.ai

When Fable 5’s cybersecurity classifiers flag a prompt — even a benign one — it automatically reroutes the query to the less capable Opus 4.8… users paying for Fable’s premium rate receive Opus-tier results due to over-sensitive safety triggers.

Hacker News commenters (via hackyournews.com) hackyournews.com

Critics argued that the term ‘frontier’ should be strictly reserved for the absolute state-of-the-art models from major labs… models like Gemma 4 or Qwen 3.6 are leaders in their respective ‘weight classes,’ [but] do not yet match the raw reasoning power of the largest proprietary systems.

Taylor Arndt, ‘Taylor’s Teardowns: Nativ’ (Substack) taylorarndt.substack.com

While the technical foundation is strong, the UI requires refinement to meet standard macOS accessibility guidelines… a ‘critical issue’ that needs to be addressed early in development.

Neura.market coverage of Nativ manifesto neura.market

Many existing apps are essentially closed-source wrappers built on top of open-source inference engines (like llama.cpp) that the developers do not actually own… hide their inner workings behind paywalls or ‘vibe-coded’ landing pages while profiting from community-driven research.

Towards AI benchmark (‘I Tested Ollama vs LM Studio on the Same Mac’) pub.towardsai.net

LM Studio’s MLX implementation was approximately 46% faster than Ollama in token generation while drawing 82% less power per token… Ollama 0.19 replaced its Metal backend with MLX, decode speeds nearly doubling — from 58 to 112 tokens per second on an M5 Max.

Michael Hannecke, ‘llama.cpp vs MLX on Apple M-series’ (Medium) medium.com

MLX-VLM faces stability issues, including occasional crashes and missing prompt caching features that can lead to inconsistent performance in long-form interactions… llama.cpp remains more mature for general-purpose inference tasks.

Kotrotsos, ‘The Local AI Stack for Apple Silicon’ (Medium) kotrotsos.medium.com

Canuma is co-founder of Neywa Labs, an AI startup that secured seed funding in June 2026 from Atomico and Seedcamp… supporting his vision of ‘on-device AI for global accessibility,’ specifically targeting regions like Africa where unreliable internet makes cloud-based AI impractical.

Edstem — ‘A Deep Dive into NVIDIA Newton’ edstem.com

Genesis reportedly achieved 8.76M FPS compared to Newton’s 196K FPS in a standard robotic arm benchmark … a CPU-bound kernel-launch bottleneck where the GPU finishes tasks quickly but the CPU fails to dispatch new work at a matching pace.

GitHub — yusufdxb/physx-newton-bench github.com

Newton backend leveraging MuJoCo-Warp can achieve nearly 2x the pure-stepping throughput of the traditional PhysX backend on consumer-grade Blackwell GPUs … End-to-end RL training saw a 1.64x speedup, while VRAM usage dropped by over 500% in some locomotion tasks.

vnrobo.com — ‘MuJoCo vs Isaac: A Deep Dive’ vnrobo.com

Isaac Sim suffers from high single-environment overhead—sometimes up to 20x slower than MuJoCo—and remains resource-heavy due to its mandatory RTX rendering and OpenUSD scene format … MuJoCo’s soft-contact model remains the ‘gold standard’ for dexterous manipulation.

Aicadium.ai — ‘The Real Bottleneck in Physical AI’ aicadium.ai

Robots trained in simulation ‘see’ everything but ‘feel’ nothing … simulators often use simplified coefficients that ignore real-world ‘stiction’ and unpredictable surface micro-textures, causing robots to drop objects that were easily grasped in simulation.

Suh et al., MIT CSAIL — ‘Do Differentiable Simulators Give Better Policy Gradients?’ groups.csail.mit.edu

In tasks like ‘Pinball’ or ‘Bounce’, gradients may only exist when a collision occurs … some simulators have been found to produce gradients in the opposite direction of the actual global optimum in highly dynamic tasks like flipping or swinging.

GitHub — google-deepmind/mujoco_playground github.com

mjlab demonstrated the ability to converge locomotion tasks for the Unitree G1 humanoid in approximately two minutes on an NVIDIA RTX 5090 … Boston Dynamics confirmed that it utilizes mjlab to train the control policies for its new electric Atlas humanoid.