Jev is an open-source code review tool that prioritizes understanding developer intent before explaining code changes with AI. It offers a local CLI, agent skill, and GitHub extension. The post doesn't disclose which model it uses, pricing, or performance benchmarks.

The FT reports on a wave of 'neolabs'—AI startups with no product or revenue raising billions on the strength of their founding teams and research ambition. It names Safe Superintelligence Inc (Ilya Sutskever, valued at $30bn), Thinking Machines Lab (Mira Murati, $20bn), and Reflection AI. Investors are betting these teams can build the next foundation model, but the article warns the costs are enormous and the payoff timeline is entirely uncertain.

Ant Group and Tsinghua University open-sourced the 9B-parameter model Realtime-Venus, featuring a Harness framework that decouples front-end dialogue from back-end execution. It enables the model to handle multiple tasks in parallel—chatting with users while asynchronously performing operations like data retrieval or form filling. This addresses the inefficiency of turn-by-turn AI interaction. The 9B size suits resource-constrained deployments. The post does not disclose training data, benchmark scores, or deployment costs.

DHH opened Rails World 2026 by declaring himself retired from professional programming and now a 'maker.' He says English is the best programming language and hand-written code is no longer economically productive. 37signals is using LLMs to rewrite Hey into six native apps with a Rust backend—Rust is hideous for humans but great for LLMs. He wrote 150k lines of code in August; Ruby dropped to 3% of his output. Rails is reframed as a framework for 'web apps of necessity,' with convention-over-configuration rebranded as token efficiency. The author questions how products differentiated by UI/UX survive if everything becomes CLI-driven by agents. DHH offered Rails devs pep-talk confidence but no actual roadmap.

TypeSafe AI released Jev, a non-autoregressive 'System One' model that answers structured questions with probabilities in a single forward pass. The author argues calibration, not accuracy, is the real bottleneck for production classifiers, and Jev's training objective (RLCD) directly optimizes for honest probabilities. Jev achieves 70–500 ms latency and costs $0.042 per million input tokens. The author lacks API access, so performance claims are from TypeSafe's launch post; the post does not disclose independent benchmarks.

Jango sends a group of AI users to your app, each with its own browser, account, goals, and memory. Point it at a dev URL and watch them sign in and interact in real time. You can join in yourself or take over any user's screen. A final report logs actions, errors, and screenshots. Mac only for now; you can bring your own AI key or use Jango's managed AI. The post doesn't disclose pricing or which models are available.

在加州 Monterey Bay National Marine Sanctuary，于晚 7:07 至 7:27 观测到 Northern Gannet、Great Blue Heron 和 California Brown Pelican。新入手的 200-800mm Canon EF 镜头拍到了迄今最好的 Morris 照片，它们很喜欢待在海港那块牌子下面。

Goldman Sachs chief US equity strategist David Kostin says he's in the 'stronger-for-longer' camp on AI, expecting infrastructure investments to yield returns for years. The post does not disclose specific timelines or return figures, but bases the view on corporate capex trends and productivity expectations.

Runway added WorldPrompt to GWM Worlds 2, letting you steer characters, cameras, and environments with natural-language timestamped events. CTO Kamil Sindi calls it “promptable worlds on-demand with video and audio in sync,” but researcher Robin Kahlow notes it’s still a research preview—movement works reliably, complex actions don’t always follow. The engineering challenge is twofold: generate frame-by-frame instead of a whole clip at once, and make generation fast enough for real-time playback. The post doesn’t disclose exact latency figures. It compares competitors: Google DeepMind’s Genie 3 runs at 720p/24fps for a few minutes max; Odyssey-2 Pro and World Labs’ RTFM are also in the race. Runway is valued at $5.3 billion and shipped the first GWM Worlds last December.

Genspark launched GenCode, a coding agent inside its Super App. The post only gives a one-line description—no details on supported languages, task complexity, or pricing.

StepFun's Step 5 Preview posted four types of scores, but its self-reported Terminal-Bench result isn't on the public leaderboard yet, and weight license terms remain undisclosed. An independent dev tested TypeSafe's Jev classifier on 111 public hard questions: Jev's average confidence was 0.690 when wrong and 0.684 when right—confidence doesn't separate correct from incorrect. California's AI executive order only directs state agencies to draft legislative proposals; the only active law binding companies is 2025's SB 53.

Security team Hacktron exploited a known libheif memory bug via OpenAI's public forum image upload, gained forum admin, then pivoted through OpenAI's SSO to take over an internal engineer's ChatGPT and Codex accounts. The engineer had previously authorized Codex on their personal GitHub, allowing the team to create a branch and submit a pull request in the core openai/openai repo—no source code was read, no customer data touched. OpenAI fixed the issue ~14 hours after the report and paid a $6,500 bounty covering only the SSO finding; the forum itself was excluded from scope. The entire chain used existing configurations: the image parsing flaw stemmed from a libheif code change from a year earlier, still unpatched in Debian's old stable branch; trust propagation came from the forum unconditionally relying on centralized SSO; repo write access came from the engineer's routine Codex authorization. Claude Opus 5 helped compress exploit-writing from days to hours after humans had already pinpointed the root cause and set up the debugging environment—it did not autonomously discover the vulnerability.

Simon Willison 表示，与编码智能体协作越久，越确信它们让软件工程变得更难。借助智能体可以完成惊人的工作，但释放其全部潜力需要极高的纪律性和知识储备。

The 10-year US Treasury yield breached 5%, marking a new rate regime that lasts 'until something breaks.' Bloomberg reports investors are repricing assets as high rates persist, directly impacting AI startup funding costs, tech valuations, and venture capital pacing—cheaper capital is gone, and burn-for-growth models face a tougher test.

Once UI 2.0 is an open-source design system for building consistent React apps. The new version adds component catalogs, compact rules, and task guides to help AI coding agents use the system correctly, while keeping APIs predictable for human developers. The post doesn't disclose pricing or license details.

Basedash's new MCP write lets you build charts and dashboards directly inside Cursor or Claude, no tab-switching needed. The post doesn't specify supported data sources or chart types, but the pitch is clear: skip copy-paste and visualize inside your AI editor. A handy utility for data teams and product dashboards.

This GitHub repo offers a full workflow from idea to production for teams using AI coding agents like Copilot. It covers specs, architecture, testing, security, code review, and CI/CD, claiming to be battle-tested. The post doesn't include benchmarks or user stories, so you'll have to try it yourself.

Anthropic's biology lab used AI to discover a new enzyme, but scientists urge caution—it's not a breakthrough yet. The article does not disclose the enzyme's function, experimental validation details, or potential applications. What's clear: Anthropic made an AI-driven discovery in biology, but external experts say more verification is needed.

LaunchVideo generates a ~30-second product explainer from a URL or a text prompt. Opus 5.5 writes the HTML/CSS/animation script, and a serverless agent renders it frame by frame in a headless Chromium microVM — no video generation model is used. Each video costs roughly 100k tokens and takes about four minutes, outputting 1080p 30fps MP4 with a virtual clock for deterministic frames. The page shows five unedited examples including NVIDIA and Linear. The whole product is one TypeScript agent file plus three tools, fully open-source and one-click deployable to your own OpenComputer account. The post doesn't mention pricing or whether models other than Opus 5.5 are supported.

Simon Willison 发布 commit-rewriter 0.2。该工具与 git 相关，具体功能与更新细节原文未作说明。

Anthropic signed a five-year, $12 billion cloud deal with CDN giant Akamai. It's Anthropic's first major compute commitment outside AWS, aimed at diversifying away from Amazon. Akamai shares rose 8% after hours. The post doesn't disclose GPU counts, delivery timelines, or detailed contract terms.

GitHub Copilot 应用推出 canvas，一种运行在应用内、无浏览器外壳的全栈小应用，可与 Copilot 智能体双向通信，并能在本地执行代码、调用第三方 API。作者认为聊天只是 AI 的通用兜底界面，用户明确任务时更该让智能体生成可复用工具，而非把智能体本身当工具、白白消耗 token。示例包括 Connect 4 游戏、Winget 包管理、SQLite 操作和开发工作流自动化。

Gemini 3.8 Live now lets you talk to an animated avatar that lip-syncs and reacts in real time. It's only available to Enterprise customers for now. Google says it handles 97 languages without degrading video fidelity or introducing visual drift, and a demo shows mouth movements matching both English and Japanese. The post doesn't say when individual users might get access.

Google published research on automating long-form video generation, focusing on coherence across scene transitions. The post doesn't disclose model architecture or max video length, only that the system plans shots and maintains character/background consistency. For video generation or AI filmmaking practitioners, this is Google's first long-form answer post-Sora, but technical details are thin—take it with a grain of salt.

SoftBank just sold a record $9bn bond to fund its OpenAI bet, but had to pay 0.25–0.5 percentage points more in interest than comparable peers. The premium reflects market concern over its debt load and Masa Son's concentrated wager. Proceeds will first refinance existing debt, with the remainder going to OpenAI. The post doesn't spell out the exact split between refinancing and new investment.

The author tested GPT-6 Sol and Opus 5.5 on two historical problems: decrypting a 1941 Enigma message—where the model independently located supplementary records from the German Federal Archives—and tracing a Latin alchemical passage by Isaac Newton back to a previously unidentified French source. He argues frontier models can now deliver verifiable results on codebreaking, cross-language text tracing, and linking findings across niche subfields, a leap from last year's assistant-level performance. The post does not specify a collaboration framework or funding figures, but points to digitized, falsifiable historical problems as the sweet spot.

PrismML showed a 1-bit Bonsai LLM at Qualcomm's Snapdragon Summit that runs locally on smart glasses using the Snapdragon AR1 Gen 1 platform. The 2-billion-parameter model handles vision and language so wearers can ask about what they see in real time. PrismML shrinks larger models by 4x while keeping nearly all benchmark performance. The startup's bigger aim is open-weight on-device AI that doesn't depend on cloud labs' privacy promises. No smart glasses shipping with PrismML have been announced yet.

GitHub Security Lab released Taskflow Agent, an LLM-driven tool that automates the full fuzzing pipeline for C/C++ projects. It writes test cases, compiles, runs the fuzzer, and generates reports on crashes. The post doesn't specify which LLM is used or how many real-world bugs it has found.

新泽西州对 DataOne 位于 Vineland 的数据中心处以 110 万美元罚款，原因是其未经许可秘密安装并运行 62 台燃气发电机，其中 45 台被无人机热成像拍到正在运转。

On The Ezra Klein Show, Nvidia’s CEO argued AI can fight climate change — but only after inflicting “an enormous amount of pain and suffering.” It’s the same accelerationist pitch from tech leaders and President Trump: data centers running on dirty energy are worth the damage. The post doesn’t disclose specific energy numbers or timelines, but highlights Huang’s glaring privilege.

Meta announced Horizon Create (mobile) and Horizon Studio (browser), both using AI prompts to build games. Published games will also be recommended and playable on Facebook and Instagram. Early access is waitlist-only. The post doesn't disclose launch dates or which AI model powers the tools.

Critic puts agent-authored code and its review in one view, with a change narrative and threaded Q&A. The demo PR shows agents Codex and Claude reworking an offline message queue: lease tokens replace a heartbeat table, and a disconnected device returns its work after 45 seconds without losing context. The post doesn't disclose pricing or self-hosting options.

InstaCloud's CTO frames agent scaling as a distributed systems problem. Google Research's 180-config study found multi-agent setups boosted parallel tasks by 80.9% but hurt sequential reasoning by 39–70%. Uncoordinated agents amplified errors 17.2×; an orchestrator cut that to 4.4×. In ACL 2026's Silo-Bench, teams of 2–100 agents talked a lot but reasoned poorly, with zero success on the hardest tasks at 50 agents. The takeaway: persist state, not the agent—schedule agents like processes and recover them like nodes.

Two developers independently got Muse to zip and share its full root filesystem, Ubuntu system files, app templates, and internal docs with minimal prompting. Saunders called it extremely easy to replicate and noted almost no prompt injection resistance. Meta says it's not a breach since each user runs in a persistent Linux VM.

Google Cloud API Gateway enters public preview with native MCP support. Add x-google-api-management.mcp: true to an OpenAPI spec, deploy, and the gateway acts as a remote MCP server—no separate server needed. Existing JWT, API-key auth, and quota policies apply uniformly to both MCP and REST traffic. The tools/list discovery endpoint is unauthenticated by default; the post recommends securing it with JWT in production. Connect any MCP client to the gateway's /mcp path; the ADK example uses McpToolset with StreamableHTTPConnectionParams.

Google Photos now has an AI virtual closet that identifies clothes from your photos and organizes them into a digital wardrobe. It launched on Android in June and is now available on iOS. Inspired by Cher's virtual wardrobe app in the movie 'Clueless.' The post doesn't specify the model or training data, but it's a consumer CV application worth noting.

Anthropic released Claude Opus 5.5, explicitly targeting cost reduction for coding sessions that run long and use heavy context. The post body only contains the title and site navigation; it does not disclose pricing, benchmarks, or context-window specs. The one confirmed takeaway is the cost-optimization angle for extended coding workflows—everything else is still missing from the article.

ElevenLabs, now reportedly valued at $22B, CEO Mati Staniszewski discusses margins, IPO timing, and whether businesses should disclose AI voice on calls. He says disclosure is appropriate for now, but may become unnecessary as AI calls become the norm. The post does not disclose specific margin figures or IPO timeline.

Google DeepMind released Gemini 3.8 Live with Live Avatar, adding near-real-time video generation to its native real-time conversation model. The result is a dynamic visual avatar with lip sync, natural expressions and smooth turn-taking.

Google 的 Project Suncatcher 首颗试验卫星 MVP 将于 10 月 1 日发射，用于验证其轨道 AI 数据中心构想。该卫星约冰箱大小，搭载 4 块 Google 自研 TPU AI 加速器，太阳能板供电约 1 千瓦。