Google uses AI to rewrite C in Rust

Hi there, this is your daily ☕️ Devshot.

In today's Devshot:

🦀 Google uses AI to rewrite C in Rust

🛡️ Nvidia launches AI agent safety platform

🚀 GKE pod snapshots cut model load times

🎨 Linear moves 1,000 PRs to Meta's StyleX

🦅 Swift 6.4 ships Subprocess 1.0

Plus: 🎁 7 other news you might like, 🧰 6 tools, and 📚 5 papers.

🦀 Google uses AI to rewrite C in Rust LINK

• Google used Gemini to translate the ~3,000-line C library giflib into an ABI-compatible Rust drop-in, published as open-source giflib-rs to serve as a reference for teams evaluating automated language transitions on foundational utilities.
• Engineers Bastian Kersting and Max Hils ran a three-stage loop, single-shot port, human review of FFI pointer ownership and lifetimes, then differential testing feeding failures back to the model, validated against 30 million real GIFs and 200 million fuzzer iterations.
• The Rust binary hit runtime parity with C while letting the team drop OS sandboxes for image decoding, cutting p99 tail latency, and it proved immune to CVE-2026-26740, an upstream giflib heap write, before public disclosure.

🛡️ Nvidia launches AI agent safety platform LINK

• Nvidia released the Open Agent Safety Platform, an open-source reference design combining software and hardware to contain AI agents and enforce what they can and can't do across any system running them.
• OpenShell creates a secure runtime that lets developers set agent limits on x86 and Arm, while Sentry runs on Bluefield-4 DPUs outside the agent to monitor activity and quarantine misbehaving agents.
• Anthropic will use OpenShell to build security into Claude Managed Agents, and partners already onboard include Microsoft, Oracle, Salesforce, SAP, ServiceNow, Cisco, IBM, Dell, and HPE.

🚀 GKE pod snapshots cut model load times LINK

• GKE Pod snapshots, generally available since May on clusters running 1.35.3-gke.1234000 or later, cut model startup latency by up to 89%, loading a 70B-parameter model in 37 seconds and an 8B model in 15.
• Built on gVisor and requiring GKE Sandbox, the feature checkpoints a workload's full running state, CPU and GPU memory, threads, file descriptors, root filesystem, to Cloud Storage, letting a new replica resume without re-running the model-loading initialization.
• Restores demand a matching machine series, gVisor kernel, and GPU driver, or the Pod boots normally; teams must also re-create encryption keys, refresh environment variables from a file, and reconnect external connections, since those aren't restored.

🎨 Linear moves 1,000 PRs to Meta's StyleX LINK

• Linear finished migrating its React apps from styled-components to Meta's StyleX, an effort spanning over 1,000 pull requests that engineer Kenneth Skovhus wrapped up in early August 2026 after roughly five months of work.
• Because StyleX shifts style generation to build time and emits collision-free atomic classes, Linear measured 20 to 35 percent less main-thread work on view-heavy pages and about 30 percent faster navigation on a mid-tier machine.
• Rather than migrate by hand, Skovhus wrote a deterministic codemod now past 500 PRs and around 100,000 lines, and Linear kept CSS Modules as an escape hatch for global selectors StyleX restricts.

🦅 Swift 6.4 ships Subprocess 1.0 LINK

• Swift 6.4 has landed with a stable Subprocess 1.0 library that gives developers a cross-platform API for launching external processes and capturing their output, alongside broader performance, concurrency, and interoperability improvements across the toolchain.
• The release cuts unnecessary copying with new types like non-copyable smart pointers and a copy-on-write-free dynamic buffer, adds borrow and mutate accessors, and lets tasks temporarily shield critical cleanup code from cancellation.
• Interoperability improves through transparent C++20 bridging and expanded Java support, Wasm code generated via JavaScriptKit now runs up to 40 times faster, and Swift Build becomes the default unified toolchain across all supported platforms.

Other news you might like

• Uber Separates Scaling Intent From Execution on Kubernetes Platform LINK

• Home Made CobbleDB Replaces DynamoDB at Perplexity to Cut Query Latency 5x and Reduce Cloud Storage LINK

• GDB 18.1 released LINK

• Reverse-engineering the vintage Intel 8087's tangent algorithm: more than CORDIC LINK

• Linux enthusiasts see 10-second kernel compilation times on the horizon LINK

• AWS Introduces Foreign Key Constraints in Aurora DSQL LINK

• Anthropic turns Claude into an AI marketplace with 2,000+ plugins and connectors LINK

🧰 Trending tools

Switch: open-source workspace connecting people and AI agents across Slack, Teams, Discord, or Mattermost, keeping context through handoffs while working with your existing tools. LINK

Kilo Code for JetBrains: an open-source AI coding agent for JetBrains IDEs that lets you build and iterate with flexible model choices across projects. LINK

Mastra Factory: a TypeScript framework for building AI agents with staged workflows, memory, streaming, evals, tracing, and a Studio UI for testing. LINK

Kane CLI: drives real Chrome with plain-English commands to test web apps and extract data, eliminating brittle selectors and scripts for developers, QA, and AI agents LINK

Harden: a free, local security tool that inspects AI coding agent tool calls before they run, keeping your repo and data on your machine. LINK

Arcjet: a runtime security platform for AI apps that detects prompt injection, authorizes agent tool calls, redacts sensitive data, and blocks bots and abuse. LINK

📚 Trending papers & reports

Silent quality gates can flag software as fully passing while skipping most of its safety checks, because missing test results get counted as no problem, a flaw fixed by adding an "unable to determine" verdict. LINK

Quantum software debugging gets an automated assistant that catches the silent failures experts hit most, lifting repair success from ~45% to ~75% versus a standard AI helper. LINK

Formal methods symposium proceedings collect 14 peer-reviewed papers on mathematically proving software and systems behave correctly, presented at a September 2026 event in Timişoara, Romania alongside a related scientific computing conference. LINK

Messy game code gets a research audit finding it suffers more bugs and lower maintainability than other software, with detection tools and testing fixes that exist but go largely unused by studios. LINK

Coding-assistant evaluations lean almost entirely on automated system tests rather than real developer feedback, meaning today's AI coding tools are rarely judged on whether they actually help the people using them, a review of 92 studies found. LINK

See you tomorrow for a new dose of ☕️ Devshot!