You can now run with .Claude Managed AgentsChat SDKClaude Managed Agents handles the agent loop server-side, including the model, tools, session state, and sandboxed web research. That means you can ship a Slack research bot built on Claude Managed Agents and Chat SDK, with one persistent session per thread and streamed briefs with sources.To try Claude Managed Agents with Chat SDK: Read moreWhat you getGet started: Replies render as the model writes them, over a single streamed response.Token-b

<p><strong><a href="https://embracethered.com/blog/posts/2026/breaking-claude-code-opus-5-and-automode/">Breaking Claude Code Opus 5 Auto Mode</a></strong></p>Anthropic are putting a great deal of faith in Claude Code's auto mode for protecting their coding agent users against prompt injection attacks. They recently <a href="https://simonwillison.net/2026/Aug/8/auto-mode/">made that the default</a> and have made bold claims about its effectiveness...

Five Rust-level memory optimizations to the DNS cache layout of Big Pineapple cut per-entry memory by 56%, freeing approximately 100 TB of memory across Cloudflare's fleet.

Socket researchers found 18 Chrome extensions and one Edge extension delivering a wallet drainer, credential theft, and other malicious payloads.

OpenClaw is the fastest-growing project in GitHub history. Peter Steinberger and several maintainers share what they learned in the project's first six months.The post OpenClaw went viral. Meet the maintainers building and securing it. appeared first on The GitHub Blog.

The now supports Cursor through the official adapter. The harness layer lets your application run different coding agents through the same interface, so you can switch agents without changing your application code. AI SDK harness layer@ai-sdk/harness-cursorHarnessAgentPass to :cursorHarnessAgentUnder the hood, the adapter uses to connect Cursor to through the Agent Client Protocol (ACP). @ai-sdk/harness-acpHarnessAgentSupported harnesses now include, in addition to Cursor, Claude Code, Cline, Co

The general idea is that we create a Document Picture-in-Picture window (DPIP window), and then we put HTML, CSS, and JavaScript into it.Creating Web Widgets Using the Document Picture-in-Picture API originally handwritten and published with love on CSS-Tricks. You should really get the newsletter as well.

The Deployments page now has redesigned filters that make it faster to find a specific deployment. You can: Open any project's to try the new filters.Deployments pageRead moreApply suggested filters for common searches. Type to find an option instead of scrolling through the list.Describe what you’re looking for in natural language, and Vercel automatically applies the matching filters.

Right, we are at the end of iteration one.The last thing to do is to make my life a little easier. Markdown files work perfectly fine for the music collection, but they're a bit of a faff. Mostly because I always forget the front matter structure, so to fix that, I created myself a Command Line Interface (CLI) which is a series of questions, resulting in a new item being added to the collection.I add a lot of music to my collection because I'm truly trying to get away from streaming platforms co

The AFP, FBI, and WA Police charged two men allegedly behind TeamPCP. Charlie Eriksen on why the arrest doesn't close the gap TeamPCP exposed.Category: News

ChatGPT デスクトップの内蔵ブラウザが、Web サイトの機能を AI エージェント向けのツールとして公開する WebMCP に対応しました。この記事では WebMCP に対応した TODO アプリを実装し、Codex が呼び出した Site tools の引数や結果を実行履歴から確認します。

Build custom dialogs in CKEditor 5 using HTML injection or the built-in Dialog plugin. Compare both approaches with a document management form example.

The idea of orchestrating long-running, stateful logic on top of unreliable, stateless infrastructure isn't new. We've had message queues, job runners, microservice choreographies, and full-blown workflow engines for a long time. What we never had was a version of it that felt good to write.I'd spent about six months working on a fork of Temporal, mostly on weekends, trying to turn it into a serverless answer with DX that felt more Vercel-native. Eventually it dawned on me that to ship that expe

Dev Machine Guard now inventories where developer tools keep credentials. See which credential sources are in use across your fleet, how many devices each one affects, and how much of that material is sitting in plaintext. Credential values, fragments, hashes, and fingerprints are never stored, displayed, or sent off the device.

Dev Machine Guard now inventories browser extensions across your developer fleet. See every extension installed in Chrome, Edge, and Firefox, what each one is currently permitted to do, whether it came from a marketplace or was installed some other way, and which devices are running it.

is now available on AI Gateway and free to use through September 25. Ling 3.0 Flash Fin from Inclusion AILing 3.0 Flash Fin is a finance-focused version of . It has a 256K-token context window, produces up to 32K output tokens, and supports reasoning and function calling.Ling 3.0 FlashThe model is designed for financial research and analysis, including multi-step workflows that user multiple tool calls before producing an answer. Choose a model ID based on what should happen after the free perio

How to add Rust and the Topcoat web framework to an existing pnpm and Nx monorepo: one task graph, shared caching, and distributed CI across both stacks.

<p><strong><a href="https://qwen.ai/blog?id=qwen3.8-flash-next">Qwen3.8-Flash-Next</a></strong></p>Another open weights model from Qwen. This one is "a multimodal MoE model that also serves as an early preview of the architecture used in Qwen4".</p><p>It's pretty big: 125B tokens, but only 6B active which means it gets a significant performance boost.</p><p>I've been trying it out on a DGX Spark using <a href="https://huggingface.co...

Safari Technology Preview Release 251 is now available for download for macOS Golden Gate and macOS Tahoe.

Managing library updates can be tedious at times. Learn how the GitHub Copilot app can handle this type of repetitive task.The post GitHub Copilot app for Beginners: Automate Dependabot pull request triage appeared first on The GitHub Blog.

Create ClickUp tasks from Socket alerts, automate ticketing with custom rules, and keep alert and task status synchronized.

Minimus is shutting down. Compare Aikido, Chainguard, Docker Hardened Images, RapidFort, and Echo, and pick a hardened-image replacement that won't lock you in again.Category: DevSec Tools & Comparisons

The CSS animation-trigger property allows you to delay the start of a CSS animation until a specific trigger occurs.animation-trigger originally handwritten and published with love on CSS-Tricks. You should really get the newsletter as well.

Rspack 2.2 is now available, featuring performance and HMR improvements, shorter module IDs, import.meta enhancements, and support for more platforms.

Data visualisation sits at the intersection of two disciplines that rarely talk to each other: data and design. Meriem Benhabiles explores what changes when you bring structured UX thinking to dashboards and data presentations, from the questions you ask before opening any tool to the decisions that determine whether an insight actually lands.

You can always easily style :hover, but what if you want the NEXT item to have styling too? Or, just a smidge harder, the PREVIOUS item. How about multiple in any direction?

Most software runs on decisions nobody made. We talk about why gating, pinning, backporting, and SBOM upkeep only work if someone actually owns them.Category: News

Aikido Security has achieved ISO 42001 certification, the global standard for AI governance, covering AI pentesting, Code Security Audit, and Deep PR Review.Category: Compliance

Aikido's agents pentest your Android app and its backend in a single whitebox assessment. You get reproducible findings, AutoFix, and retests for every issue.Category: Product & Company Updates

<blockquote cite="https://pauldix.com/the-end-of-programming"><p>The fact that AI wrote 1M LOC and then refined it over the course of the next couple of months to produce a reliable piece of software that is currently running on millions of developer machines is absolutely mind blowing. And you can say, “well it’s not that impressive because they had an oracle to compare against, so it was simple to go from one language to another”, but I think that’s selling this entire thing short...

本記事は、LINEヤフーの技術カンファレンス「Tech-Verse 2026」のセッション「10x Speed With QA Agent Platform — How we scaled adopt...

LINEヤフーでは、ユーザーの目的に応じてさまざまなサービスや機能をつなぐAIエージェントサービス「Agent i」の開発を進めています。Agent iで多様なニーズに応えるには、現場のアイデアを素早...

こんにちは。株式会社タップルでエンジニアリングマネージャーをしている山岸です。 この度、2026年6 ...

The SF Ruby Startup Conference is an engineers' conference: meet the people building Ruby open source, scaling Rails in production, and shaping what agentic Rails looks like next.

pnpm 12 is stable. It is a rewrite of pnpm in Rust, and it is deliberately not a migration: the commands, flags, settings, and lockfile format of pnpm 11 all carry over, and the documentation describes both versions.

from Meta Superintelligence Labs is now available on AI Gateway. It is their first image model and a separate family from Muse Spark, returning images rather than text. Send a prompt and get an image back, or send an image with an instruction and get it changed. One model does both, so you don't switch models to move from generating to editing.Muse ImageTo use Muse Image, set to and call from the :modelmeta/muse-image-1.0generateImageAI SDKTo steer the result toward art you already have, pass re

from Google is now available on AI Gateway for recorded and live audio:Gemini 3.5 TranscribeBoth models automatically detect more than 85 languages, including when a speaker switches languages. You can also provide custom vocabulary to improve the transcription of names, technical terms, and uncommon spellings. The model for complete recordings can also identify speakers and return word-level timestamps.Streaming transcription is available in . Install the latest AI SDK and AI Gateway provider:

from Alibaba is now available on AI Gateway. It takes text and images as input, serves a context window of 1 million tokens, and can return up to 65k tokens in a response. Alibaba recommends it for coding, tool use, and multi-step agent workflows.Qwen 3.8 FlashTo use Qwen3.8-Flash, set to in the :modelalibaba/qwen3.8-flashAI SDKTo use it in a coding agent, see the , then run to connect agents like Claude Code, Codex, OpenCode, Cursor, Pi, and more and select inside the agent.coding agents guidev

is now available on AI Gateway as .GLM 5.3 Flash from Z.aizai/glm-5.3-flashGLM 5.3 Flash is a multimodal coding model with a 1M-token context window. It accepts both text and image inputs and supports function calling, structured output, and streaming.To include images, pass them with text in a message. A request can include multiple images using URLs, Base64 data URLs, or binary data: To use GLM 5.3 Flash with a coding agent, run , then select as your agent's model.vercel ai-gateway coding-agen

The Vercel Security Dashboard is now generally available on all plans, giving you one place to see your security posture across every account and project.You can access the Security Dashboard in or run in the Vercel CLI. the UIvercel security checkAs teams grow and coding agents make it faster to spin up projects, small misconfigurations add up quietly. The Security Dashboard automatically flags issues like:Misconfigurations are ordered by risk with the most severe first, and each finding links

Python projects can now use to set response headers or rewrite requests to internal paths, including apps built with FastAPI, Django, and Flask.routing rulesThe evaluates rules before requests reach your application, so changes apply without a new deployment.Vercel CDNFor example, this FastAPI app serves a route:/newTo send requests for to that route, create a rewrite from the CDN tab in your project dashboard or with the :/oldVercel CLIPublished rules take effect immediately across all regions,

<p><strong><a href="https://www.eveonline.com/news/view/the-move-to-python-3-begins">EVE Online: The Move to Python 3 Begins!</a></strong></p>EVE Online has been one of the most interesting case studies in Python at scale for over twenty years now.</p><p>They've been running on <a href="https://github.com/stackless-dev/stackless/wiki/">Stackless Python</a> since their launch in 2003, and their last major upgrade was 16 years ago, to St...

These are the lessons we learned evaluating LLMs for real-world secret scanning.The post How to evaluate LLMs before production appeared first on The GitHub Blog.

Create and manage Asana tasks directly from Socket alerts, with manual task creation, automated ticketing rules, and two-way sync.

SummaryImpact on Vercel deploymentsResolution for self-hosted applicationsCreditReferencesTwo vulnerabilities affecting Next.js were disclosed in the August 2026 Security Release. Next.js applications hosted on Vercel are protected and require no customer action. Next.js disclosed the following critical vulnerabilities: After the AVIF vulnerability was identified, Vercel applied protections to its managed Image Optimization service.Applications hosted on Vercel are protected. No upgrades, config

now has a free tier that gives you a high-level performance overview from your real users. The new free tier: Speed InsightsInstall Speed Insights: Previously, free Speed Insights was limited to a single project on Hobby, and upgrading to Pro meant losing access unless you paid for the add-on. Now your free tier carries over, and you only pay if you upgrade.The paid product, now called Speed Insights Plus, includes deeper diagnostics and historical data. If you were already paying for Speed Insi

A brand new 2 KB client-side syntax highlighter from Dave. Includes a web component for very easy usage. Maybe these web components have legs, eh? The most interesting part about it is that it doesn’t touch the DOM. It uses the Highlight API to colorize text and TextMate grammars (the same as in VS Code) […]

Syntax highlighting for code blocks without the complicated markup, spans, classes, and bloated JavaScript, courtesy of Uncle Dave.MicroLighter: Syntax Highlighter originally handwritten and published with love on CSS-Tricks. You should really get the newsletter as well.

We recreated the viral AI gym hack in a controlled environment. Running Opus 4.6 on OpenClaw, the model exploited the booking flaw in nine of ten runs.Category: Research

Every website peaks on launch day and slowly drifts from there, not because it breaks, but because nobody has time to keep it current. Autonomous websites, continuously optimized by agents after launch, aim to change that. Pierre Burgy shares what they learned building for full website autonomy and the deeper design problem they uncovered along the way.

IndexedDB は、構造化されたデータをブラウザへ保存できる非同期・トランザクション型のデータベースです。この記事では素の IndexedDB API で基本的な仕組みを確認した後、React と Dexie.js でタスク管理アプリを実装します。

Agents increasingly write TypeScript programs to coordinate tools and process their results. Once those programs touch real applications, some steps require authentication, while others need human approval.Executing that code with gives it the same access as the application around it, including its secrets and internal services, and leaves no durable way to pause at those boundaries.evalToday, we're releasing the , a package for executing untrusted JavaScript and TypeScript without giving it dir

Every useful agent reaches beyond your codebase. It posts to Slack, opens pull requests, queries Snowflake, or calls an internal API. That reach is what makes it valuable, and it's also where the risk lives, because for years, granting it meant provisioning a long-lived token and hoping it never leaked. replaces long-lived tokens with ones your code requests at runtime, scoped to the task and expiring on their own.Vercel ConnectDuring the , we've grown the ecosystem past 100 connectors, unified

Our Rails AI assistant had read-only SQL access and could still return peer review scores. Learn how we kept open-ended analytics in an in-app AI assistant while making database access honor application permissions.

and are free on AI Gateway via GMI Cloud through Sunday, September 6.MiniMax M3M2.7Use or to route requests to GMI Cloud. These model IDs will return an error after the free period ends.minimax/minimax-m3-freeminimax/minimax-m2.7-freeTo keep requests working after the free period, use the standard model ID without the suffix and place GMI Cloud first in the provider order:-freeAI Gateway tries GMI Cloud first and can fall back to another provider if GMI Cloud can't serve the request. This lets t

is now available on AI Gateway as .Wan 3.0 from Alibabaalibaba/wan-v3.0-videoWan 3.0 combines text-to-video, image-to-video, first- and last-frame conditioning, and reference-based generation in one model. References can include images, video, and audio. It generates clips up to 30 seconds at 30 fps in 480p, 720p, or 1080p, with synchronized audio. Previously, Wan 2.7 required separate and model IDs and was limited to 15-second clips at 24 fps.-t2v-r2vWan 3.0 supports , so no HTTP request needs

on AI Gateway can now run asynchronously. Video generationBy default, keeps one HTTP request to AI Gateway open until the result is ready. Because video generation can take seconds or minutes, that request can exceed request timeouts.generateVideoWith asynchronous generation, your application can receive a webhook, poll for completion, or start a generation and retrieve the result in a later request.Choose an option based on whether your process can keep running and whether your application can

is now generally available on all plans and in . Instead of storing long-lived provider secrets, your code requests short-lived, scoped tokens at runtime. Deployments authenticate with their existing Vercel OIDC identity. Each token is scoped to the task, refreshed automatically, and expires on its own.Vercel Connectv0Register a connector . Pass the service name and the CLI pre-populates the brand name, icon, auth type, and MCP or discovery URL, then prompts for any credentials the service requi

Chat SDK's Slack adapter now supports .Slack Enterprise GridBots installed org-wide work across every workspace, with correct token resolution, tenant-scoped caches, and event retry deduplication.The adapter now stores org-wide installations by enterprise ID. This matches how tokens are resolved for incoming events, slash commands, and interactive payloads. records the new identity fields:SlackInstallationToken resolution behaves the same over HTTP webhooks and Socket Mode. Events route by the i

now includes a managed connector for , so your apps and agents can send and receive messages over iMessage, RCS, and SMS.Vercel ConnectLinqAs a , Vercel can create a Linq account and phone number for you, or link an existing account. You never manage credentials yourself.Vercel Managed ConnectorCreate a connector from the or :dashboardVercel CLIThe connector powers the new in .Linq channeleveRun , choose Vercel Connect, and eve wires up the connector, phone numbers, and webhook for you:eve add c

Your team already works in Notion. Now your agent can too.With the new for Chat SDK, the same agent you run on Slack, Discord, GitHub, Teams, or WhatsApp can join comment discussions on your Notion pages, no separate codebase required.Notion adapterEach Notion page maps to a channel and each comment thread to a thread, so replies stay threaded automatically.The adapter supports mentions, message editing, conversation history, and up to three file attachments. By default, your bot replies when @-

You can now build bots that hold end-to-end encrypted 1:1 and group conversations on XChat with the new for Chat SDK.XChat adapterThe adapter handles all encryption, key management, and signature verification automatically. Bots can also message users first, as long as the user has encrypted chat set up and follows the bot.XChat has no markdown rendering, so the adapter falls back automatically: URLs and @mentions render as tappable links, tables as ASCII code blocks, and cards as text with a li

Hi, I’m a backend engineer working on Bucket ...

はじめに カミナシでSWEをしている osuzu です。 筆者はSaaSと呼ばれる業務ソフトウェア開発を前職含め6年ちょっと経験し、最近ようやく言語化できたことがあります。 それは「SaaSの技術的な難しさは、ホリゾンタルかバーティカルかで種類が異なる」ということ。そしてより重要なのは、この2つの戦いを混ぜてしまうと、どちらの戦い方も難しくなるということです。 前提: ホリゾンタルSaaSとバーティカルSaaS 念のため用語を揃えておきます。 ホリゾンタルSaaS: 業界を問わず横断的に使われるSaaS。チャット、タスク管理、通知、ドキュメントなど。SlackやNotion、Google Do…

We built a plugin for the GitHub Accessibility Scanner to make sure your alt text is actually accessible. Here's how it works.The post Your alt text passes automated checks. That doesn’t mean it’s any good. appeared first on The GitHub Blog.

We migrated the Cloudflare Blog to EmDash to prove our stack at massive scale. Here is how we stress-tested performance, safely routed production traffic, and redesigned the frontend experience.

<p><strong>Release:</strong> <a href="https://github.com/simonw/llm-anthropic/releases/tag/0.27">llm-anthropic 0.27</a></p> <p>This release of the Anthropic plugin for <a href="https://llm.datasette.io/">LLM</a> mainly provides compatibility with the recently released <a href="https://github.com/anthropics/anthropic-sdk-python/releases/tag/v1.0.0">anthropic v1.0.0</a> Python library, which switches from <code>httpx</code...

now consider cache hits when deciding whether to use a smaller build machine. A warm-cache build no longer triggers a downgrade. Elastic build machinesTurborepoA warm-cache build can use less CPU and memory than the same build with a cold cache. Downgrading based on that lower usage could leave a later cold-cache build without enough resources to complete successfully.This change applies automatically to all builds using Elastic build machines. No action is required. Learn more in the .build doc

It isn’t often that new image formats land in browsers. In the early 2000s we had JPEG, GIF, and PNG. The 2010s gave us WebP, which was a modest step up from JPEG. But the 2020s have given us two new image formats that are a big step up from previous formats: AVIF and JPEG […]The post Intent to Ship: JPEG XL appeared first on Mozilla Hacks - the Web developer blog.

記事は ics.media へアクセスしてご覧ください。

CKEditor AI now works from your own prompts and files, fits what your editor is set up to do, and understands the images in your document. Table editing and formatting around widgets are also improved.

Seven and half years after blocks arrived in Core, WordPress introduces a way to build blocks without React annd build pipelines. All you need is PHP.WordPress PHP-Only Block Registration originally handwritten and published with love on CSS-Tricks. You should really get the newsletter as well.

My post Thinking Horizontally in CSS @layer got Stuart Robson thinking about how to pull it off at a bigger scale. Stu is right that our components are essentially hand-written, whereas with a more enterprise-scale design system, design tokens are more likely to come from a build pipeline. That got Dave Rupert thinking about how […]

"Let's say you have 4 issues that can be implemented simultaneously..."

<p><strong><a href="https://fzakaria.com/2026/08/23/your-executable-is-a-sqlite-database">Your executable is a SQLite database</a></strong></p>Farid Zakaria describes a neat Linux pattern for creating a SQLite database file that can be directly used as an executable binary.</p><p>The trick sets the SQLite file format's 4-byte application ID (68 bytes into the file) to SELF, standing for Structured Executable & Linkable Format. The various ...

Aikido AI Code Audit found 8 more vulnerabilities than Claude Security with Mythos at less than half the cost. How harness design drives coverage per dollar.Category: Research

はじめに こんにちは！2026年7月の1か月間、「CA Tech JOB」に参加した池畑です！ 株式 ...

npm Trusted Publishing sat near-idle after it was released. Then Shai-Hulud and 14 more supply chain attacks pushed adoption 3.4x. Charlie looks at the data behind it.Category: News

Open VSX has removed three extension IDs from its malicious-extension list as the legitimate publishers they impersonated move to claim the names for themselves.

now runs globally, starting with four regions: (Washington, D.C.), (San Francisco), (Cleveland), and (Paris). Vercel Sandboxiad1sfo1cle1cdg1 remains the default. Support for all Vercel regions is coming soon.iad1Choose a region close to the databases, object storage, and other services your sandboxes access to reduce latency. Region selection is available on all plans. Pro and Enterprise teams can also configure failover regions. If the primary region is unavailable, new sandboxes start in the c

PR TIMESでCTOをやっている金子 (@catatsuy) です。 2023年に、PR TIMESのログインシステムへJWTを導入したことを開発者ブログで紹介しました。 この記事では、JWTの仕様や移行の進め方など […]

HighlightsThis patch release fixes false positives in the no-loss-of-precision rule that were introduced in v10.9.0.Bug Fixes1e641c9 fix: no-loss-of-precision false positive with trailing decimal point (#21251) (Aleksandr Shoronov)Documentationad74a8d docs: add deprecation steps for EOL package versions (#21248) (Francesco Trotta)Chores3c3ae53 chore: update ecosystem plugins (#21249) (ESLint Bot)

pnpm 11.24 brings back pnpm approve-builds --global, which isolated global installs took away in v11.0, and makes recursive batch publishing group its packages by registry so a credential mismatch is caught before anything is published. It also stops --frozen-lockfile from failing over the pnpm version the lockfile pins.

When you add or edit an environment variable in Vercel, you now choose Config or Secret instead of using the Sensitive toggle. Existing variables marked Sensitive are automatically treated as Secrets and continue to work without migration.You can select an environment or Preview branch for each value. The in the dashboard shows each variable’s type and where it applies.environment variable listThe team policy is deprecated with this update. When enabled, it required every environment variable cr

The on Vercel Functions now supports larger package sizes up to 5GB uncompressed and extended max duration for up to 30 minutes, two betas that previously ran on Node.js and Python only.Bun runtimeLarge functions raise the standard 250MB package size limit to 5GB, and extended max duration raises the generally available 800-second ceiling to 1800 seconds for Pro and Enterprise teams. Both features require to be enabled. New projects are enrolled in the large functions beta automatically, while e

Apps and agents built in v0 can now securely connect to , including Slack, Google, Notion, GitHub, and Salesforce, through .more than 100 servicesVercel ConnectAsk v0 to connect your app to a service. It opens the connector setup and guides you through it.You set up each connector once for your team, then reuse it across apps.For managed connectors like Slack and GitHub, Vercel handles app registration, so you don’t need to create or configure an app with the provider. For other services, you ca

はじめに 株式会社AJA でバックエンドエンジニアをしている片山です。 AJA DSP では、Goo ...

今年も AWS re:Invent がラスベガスで開催されますね。私は去年初めて現地参加し、今年も参加する予定です。この記事では、今まで現地参加に消極的だった私が、去年悩んだ末に初参加して感じた「現地参加すべき理由」を紹介します。もし、この記事をきっかけに現地参加を決めた人がいたら X(saramune) で連絡をください。現地でお会いできたら、ぜひ一緒にご飯でも食べながら AWS 談義をしましょう！（私が美味しいお店を探しておきます！） なお、本記事は AWS さんの連載「「ラスベガス5日間で得たもの」── re:Invent に賭けたスタートアップのリアル」にて、弊社 CTO がインタビュ…

StepSecurity threat intelligence tracked 56 open source supply chain attacks from August 2025 to August 2026, roughly one every three days since March. See the data and the defenses.

<p><strong><a href="https://www.ft.com/content/5ee49718-c258-4f01-aa32-7e5b76ae5245">Anthropic’s best AI model struggles to attract users as cheaper tools thrive</a></strong></p>A few interesting numbers in this FT story gathered from "people with knowledge of the matter":</p><ul><li>Anthropic's "annualized revenue" for July is up to $65bn - it was $47bn in May, and I collected <a href="https://simonwillison.net/2026/May/29/anthropic/"&gt...

<blockquote cite="https://www.dbreunig.com/2026/08/23/fable-the-end-of-moore-s-law.html"><p>Prior to Fable, it felt silly to waste <em>too</em> much time improving your coding harness or context strategies. A new model would arrive at the same price (or cheaper!) and paper over most of your problems.</p><p>But then Fable landed. It was (and still is!) <em>incredible</em>. But the cost was so high and Opus was <em>good enough</em> (as w...

Vitest で作成したスパイを復元し忘れると、別のテストへモックの状態が漏れるおそれがあります。Vitest 3.2.0 以降では `vi.spyOn()` の戻り値を `using` で宣言すると、スコープを抜けるときに元の実装を自動的に復元できます。この記事では `using` による自動復元の仕組みを紹介します。

pnpm 11.23 gives the registries setting a shape that describes each registry once — its tarball layout, the scopes routed to it, the prefix it answers to — so an Artifactory or GitLab registry can finally keep its tarball URLs out of pnpm-lock.yaml. It adds virtualStoreType, makes undeclared imports resolve under ESM with the global virtual store without a plugin, teaches pnpm config get to report the settings pnpm actually acts on, and warns about settings no pnpm version recognizes.

<blockquote cite="https://github.com/torvalds/linux/commit/818bebeb63dd6bf5f4e07e145f6cdbace520a34c"><p>And this was a debug session from hell, enormously helped by an AI doing much of the grunt-work.</p><p>I'd like to call it my tireless helper, but the AI several times stated flat out that this was impossible and unsolvable and that we should just write a report about it.</p><p>I suspect those things have been trained by people who may not be quite as stubb...

<p><strong>Release:</strong> <a href="https://github.com/simonw/llm/releases/tag/0.33">llm 0.33</a></p> <p>My highlights from this release:</p><blockquote><ul><li>Upgraded to the OpenAI Python library 3.x and switched the HTTP client dependency from <code>httpx</code> to <code>httpx2</code>. <a href="https://github.com/simonw/llm/issues/1608">#1608</a>, <a href="https://github.com/simonw/llm...

<p>The key skill required to make productive use of coding agents is being able to confidently instruct them on how to make changes and then confidently verify that those changes have been applied in the correct way.</p><p>Sometimes this involves reviewing every line of code they have written, but there are other ways to achieve that goal. Eyeballing every line of code has never been the most effective way to validate a change to a piece of software.</p> <p>Tags: &...

Three Rust crates are compromised: arrayref 0.3.10, internment 0.8.7, and append-only-vec 0.1.9 each added a typosquatted build-time dependency (proc-macro1, proc-macro-en) whose build script downloads and runs a remote binary during cargo build. Full technical analysis: timeline, dropper dissection, runtime detection, IOCs, and remediation.

AI が書いた記事は一般論に寄りすぎ、冗長になりやすいと考えていたため、これまでは誤字脱字の確認にだけ AI を使っていました。AI に任せる執筆フローを試したところ、生成稿を批判的に読み、自分の体験を加えて推敲すれば、学習効果は大きく失われないと感じました。この記事では、実際に試した方法について紹介します。

Slack Code は Slack 上でオープンな開発を行う手段を提供する機能です。コードチャンネルと呼ばれる特別なチャンネルを作成し、チーム全体でコーディングエージェントとやり取りを行うことができます。この記事では Slack Code を実際に試してみた感想を紹介します。

In short: Vim has a joke autocmd called UserGettingBored that doesn’t do anything.Vim’s automatic commands feature, usually shortened to “autocmd”, lets you run code when various events occur. For example, you could implement an auto-save feature by binding the TextChanged event to the :w command.Vim has over 100 events, from “buffer was created” to “file was saved”. But one of them sticks out to me: UserGettingBored. Here’s the documentation:UserGettingBored: When the user presses the same key

A new language joins the toolchain, downloads can now be verified with GPG signatures, and

Cloudflare's new Bot Preference Sync automatically aligns your robots.txt file with your AI bot policies for Search, Agent, and Training. Easily manage which bots access your content without maintaining static files.

Ora on VercelEvery harness expects its own infrastructureOne platform under every harnessTesting eve like any other harnessThe framework behind Ora's own agentsFront end, back end, and agent runtime on one platformEvery major agent tested side by side on live sitesHundreds of commits a day from a 16-person engineering team7% fewer steps to reach the goal2x native success: twice as many tasks finished on the customer's own site instead of falling back to web search9% more valid endpoints: more of

Image hardening tools compared for 2026: Aikido, Chainguard, Docker, RapidFort, Echo, Minimus, and WizCategory: DevSec Tools & Comparisons

Compare the top enterprise AI pentesting tools of 2026: Aikido, XBOW, NodeZero, Pentera, Hadrian, and Cobalt.Category: DevSec Tools & Comparisons

Socket’s PHP and Composer support is now in Beta for all customers, with PHP reachability analysis generally available.

<p><strong>Release:</strong> <a href="https://github.com/simonw/llm/releases/tag/0.32.1">llm 0.32.1</a></p> <p>Fresh installs of LLM stopped working the other day because the OpenAI Python library dropped its usage of <code>httpx</code>, and it turned out LLM depended on that library but only installed it via a transitive <code>openai</code> dependency.</p><p>This dot-release fixes that for the moment by pinning to &l...

<p><strong>Release:</strong> <a href="https://github.com/simonw/llm-openrouter/releases/tag/0.7">llm-openrouter 0.7</a></p> <p>Now that this plugin is compatible with <a href="https://simonwillison.net/2026/Aug/4/new-release-of-llm/">LLM 0.32</a> it can display the reasoning traces for LLMs available through OpenRouter.</p><blockquote><ul><li>Updated for compatibility with <a href="https://llm.datasette.io/en/stabl...

<p><strong><a href="https://sockpuppet.org/blog/2026/08/20/stop-making-tuis/">Stop Making TUIs</a></strong></p>Thomas Ptacek advocates for building real native user interfaces for even the smallest of personal tools, because coding agents have reduced the cost of getting a usable-enough GUI up and running to almost nothing.</p><p>I wrote about my vibe-coded bandwidth and GPU monitoring macOS task bar apps <a href="https://simonwillison.net/2026...

A newly resolved proposal would allow us to select classes that are a prefix for variations with a wildcard, like .prefix-*.Resolved: CSS Class Prefix Selector originally handwritten and published with love on CSS-Tricks. You should really get the newsletter as well.

<blockquote cite="https://interconnected.org/home/2026/08/21/galactic"><p>After I released version 1.0, I figured I would have to do the rotations myself. So I sat down with ChatGPT and I didn’t get it to write the code, but I got it to educate me. With a patient, interactive tutor, I was able to finally do what I hadn’t by reading books and asking mathematician friends – I learnt how to use quaternions just enough to make the app work.</p><p>So learning doesn’t stop jus...

CVE remediation is fixing known flaws in the software you run. Why upgrading often fails, what remediation actually involves, and how backporting fixes it.Category: DevSec Tools & Comparisons

Websites often hardcode keyboard shortcuts for Mac (⌘) while neglecting Windows users, leading to confusion.

Three months ago, I wrote that Bun was the only mainstream package manager that couldn't do surgical CVE remediation. Bun 1.4 ships in-place transitive updates, bun audit fix, and nested overrides. Time to re-run the original reproductions.

BulletedThis is the stuff that's exciting about the AT protocol. Not the "new twitter" bullshit, but the endless possibilities that this technology opens up. It's using the fancy new private data stuff too.The future of CSS: target multiple classes with the class prefix selectorAs Bramus says in the article, we can sort of do this already, but those substring selectors don't perform well. This new method is a very good improvement!Ruminations on notificationsA good write-up on how annoying and h

We tested 10 AI models on 32 fresh CVEs. DeepSeek V4 Pro found 28, and three cheap runs beat one pass of Opus 5 or Grok on total coverage.Category: Research