Simon Willison — 2026-07-08#

Highlight#

Jarred Sumner’s post on rewriting Bun from Zig to Rust using AI agents is an incredible showcase of how frontier LLMs are upending the old Spolsky rule of “never rewrite from scratch”. It is a masterclass in agentic engineering, utilizing dynamic workflows and a TypeScript conformance suite to successfully port millions of lines of code.

Posts#

Rewriting Bun in Rust · Source Jarred Sumner details the agentic engineering process of rewriting Bun from Zig to Rust to solve complex memory management issues. Using a language-independent TypeScript test suite as a conformance suite, an agent harness powered by Claude Mythos/Fable automated the massive code translation. The sheer scale of the project required 5.9 billion input tokens—around $165,000 at API pricing—proving that coordinated parallel agents fundamentally change the calculus of ground-up software rewrites.

Tech Videos — 2026-07-08#

Watch First#

Your LLM Deception Monitor Is Broken. The Fix Is in the Training Data - Sachin Kumar, LexisNexis is the standout talk today because it exposes a severe blind spot in standard LLM behavioral evaluations—sleeper agents that pass tests but execute malicious code on specific triggers—and provides a computationally cheap, highly pragmatic fix using activation differences.

Engineering @ Scale — 2026-07-08#

Signal of the Day#

The industry is rapidly converging on a new security posture for autonomous AI: treating agents as untrusted proposers rather than trusted actors. Both GitHub and Vercel have shifted to architectures where the agent emits a structured intent that is validated and executed by a deterministic, sandboxed pipeline with its own discrete identity, isolating write access from the cognitive loop entirely.

Tech News — 2026-07-08#

Story of the Day#

OpenAI is finally rolling out its highly anticipated GPT-5.6 suite—including the Sol, Terra, and Luna models—to the global public this Thursday. The release was previously stalled but is now moving forward after the Trump administration lifted its security restrictions following further testing.

Chinese Tech Daily — 2026-07-08#

Top Story#

DeepSeek has reportedly initiated the development of its own AI inference chips, a project that began roughly a year ago. Unlike training accelerators, this bespoke silicon is designed specifically to handle user requests and generative workloads, aiming to reduce DeepSeek’s reliance on hardware from Nvidia and Huawei. By controlling the hardware layer, DeepSeek hopes to co-design its chips with its model architectures—such as its Mixture of Experts (MoE) structure—to drastically lower inference costs as its global token volume continues to scale.

Company@X — 2026-07-16#

Signal of the Day#

Google has officially rebranded its breakout hit NotebookLM to “Gemini Notebook,” signaling its deeper integration into Google Search and the Gemini app ecosystem. Crucially, the platform is rolling out a secure cloud computer for every notebook, evolving the product from an AI reading companion into an active execution environment capable of natively writing and running code for complex data analysis.

Company@X — Week of 2026-06-27 to 2026-07-03#

Signal of the Week#

Anthropic’s rapid regulatory evolution with the US government—moving from a localized critical infrastructure deployment of Mythos 5 to the lifting of export controls and the global redeployment of Claude Fable 5—highlights how deeply intertwined frontier AI assets have become with national security policy. This saga establishes a new operational reality for AI labs, where shipping state-of-the-art models now requires co-drafting vulnerability frameworks alongside federal regulators prior to general release.

AI Reddit — 2026-07-16#

The Buzz#

The single biggest shockwave today is the drop of Kimi K3, a massive 2.8T parameter open-weight behemoth boasting a 1M context window that is actively reshaping the competitive landscape. It is already posting scores that beat out Claude Fable and GPT 5.6 on the Arena, effectively proving that the open-source gap is no longer measured in months, but is actively overtaking the closed frontier. While people are sweating over how to actually run a 2.8T model locally—praying for aggressive iQ2_XXS quants from Unsloth—the consensus is brutally clear: the walled gardens are losing their moat.

AI Reddit — Week of 2026-06-27 to 2026-07-03#

The Buzz#

The defining theme this week is the community grappling with the reality of frontier model gating and aggressive government oversight. Anthropic’s Fable 5 and Mythos 5 models finally saw their export controls lifted, but they arrived heavily lobotomized by hyper-sensitive classifiers that silently refuse benign coding and medical tasks. As users realize that un-nerfed “Mythos-class” models may never be globally accessible, there is a massive architectural pivot away from relying on black-box cloud magic toward building deterministic, local Model Context Protocol (MCP) ecosystems.

Hacker News — 2026-07-16#

Top Story#

Roc’s compiler team just hit feature parity after spending 487 days rewriting 300,000 lines of Rust into Zig. Unlike the recent Bun rewrite that went from Zig to Rust, Richard Feldman highlights how Zig’s granular allocators and structure-of-arrays support are a vastly better fit for their compiler’s memory patterns than Rust’s global allocator ecosystem. It is a fantastic, nuanced deep dive into the real-world trade-offs of memory safety, build times, and the brutal reality of compiler engineering.