Are you using vLLM?

Anthropic’s Project Glasswing—restricting Claude Mythos to security researchers—sounds necessary to me Anthropic didn’t release their latest model, Claude Mythos (system card PDF), today. They have instead made it available to a very restricted set of preview partners under their newly announced …

Wrote up some thoughts on Anthropic's Project Glassing, where their latest Opus-beating model is available to partnered security research organizations only. Given the recent alarm bells raised by credible security voices I think this is a justified decision.
simonwillison.net/2026/Apr/7/p...

Now if I only had a MacBook with 128Gb RAM.

New blog post: www.yduman.dev/posts/local-...

My prediction towards the end of 2025 was that 2026 will be the year where enterprise will adopt AI hard. It's happening. Enterprises are now hunting for skills that barely anyone has. Bootstrapping AI infra in this space comes with its own challenges.

JSHeroes 2026 | Community Organized JS Conference JSHeroes is an yearly event organized by the local JS community in Cluj, Romania. Check the latest updates and join us on the 14th and 15th of May 2026

I love URL text fragments but I can never remember if it is ~:~ or :~: and that I have to type text and that I need a hash as well and, well, ... basically all of the syntax. Anyway, I'll be at JSHeroes in May, talking about forensic techniques for frontends :) jsheroes.io#:~:text=Fore...

A screenshot of the copilot terms of use, available online at: https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse The terms "advertising" and "entertainment" purposes only are highlighted.

I love the Copilot terms of use. Advertising, processing of data, Copilot is not responsible, it is only for entertainment, and Microsoft is not at fault.
www.microsoft.com/en-us/micros...

I have a talk about the „renaissance of agile“ in our Ai world that argues that maybe we should still focus on building the right thing over building the wrong thing fast. Seems something that we could benefit from.

.... then I'll humbly suggest to add linters, architecture tests (dependency cruiser, TsArch, ArchUnit), duplicate code finders, and generally more static code analysis (CodeScene, Quodana, SonarQube).

Glibness aside, that is actually how you do it. Genies are quite bad at first drafts but pretty good at critiquing code. If you want more determinism (and faster generation times and less token waste),

A flow chart diagram that goes from original prompt to generated code to a decision with the question: "Are you sure it's ok?". If the coding genie decides "no", it creates a fix prompt that will lead into the same loop again. Only leave that loop if the genie says yes. If you ask "are you sure it's ok?" between 10 and 100 times, the genie will generate good code.

Breaking. I have scanned the whole web and distilled the knowledge of how we get coding genies to write good code into one image. Behold.

How do worktrees solve merge conflicts. It seems they only delay them, no?

😞 I am sorry. For people. For you having to deal with people. Thank you for publishing the advisory.

while true {
generate code over there;
}

Seems solvable.

Wow. That was a fast article. Didn’t you read about this just the other day?

Sora on X: "We’re saying goodbye to Sora. To everyone who created with Sora, shared it, and built community around it: thank you. What you made with Sora mattered, and we know this news is disappointing. We’ll share more soon, including timelines for the app and API and details on" / X We’re saying goodbye to Sora. To everyone who created with Sora, shared it, and built community around it: thank you. What you made with Sora mattered, and we know this news is disappointing. We’ll share more soon, including timelines for the app and API and details on

Link to the Sora denouncement: x.com/soraofficial...

Not because we like it but because we have found that to work...better.

Focused review agents for a11y, architecture, behaviour, complexity, concurrency, delivery pipelines, documentation, duplication, domain, infrastructure, naming, performance, security, spec compliance, structure, test quality, UX, etc. Running in parallel, burning tokens, fixing issues.

For coding agents they have just gone up. Instead of simple queries we now construct multi-agent workflows with context clearing in-between. To ensure quality code we launch multiple review agents focused on different aspects of quality.

OpenAi is shutting down Sora. I guess they realized that the unit economics of video generation are horrible. Maybe they also realized that Sora would not convert more free users to paying users. Or maybe they are focusing on areas where the unit economics are better.

But where are they better?

Wow. It’s been five years already. Kudos for making great tooling.

Without further comment.

opencode 1.3.0 will no longer autoload the claude max plugin

we did our best to convince anthropic to support developer choice but they sent lawyers

it's your right to access services however you wish but it is also their right to block whoever they want 1/2

they literally renamed the entire company after this

Today is my last day with the Deno team 🦕💖💔

I know they're gonna keep making awesome things.

But now *I* need to make awesome things for someone else! If you're looking for a DevRel with a JS focus and extra sparkle, get in touch!

🤪

I still think this is a strong contender for the funniest video ever uploaded. Absolutely unbeatable comedic rhythm

Did you hear the one where Java had an emergency release of the JDK because the year in Japan is determined by the acting emperor and guess which country just lost its emperor?

Yep. Not looking forward to it.

Bypassing AWS Bedrock AgentCore Sandbox via DNS | BeyondTrust Phantom Labs research reveals AWS Bedrock AgentCore Code Interpreter sandbox mode allows DNS queries. This flaw enables network isolation bypass.

The ways to escape a sandbox are quite glorious:
> AWS Bedrock AgentCore Code Interpreter’s ‘Sandbox’ mode allows DNS queries, enabling *interactive shells* and bypass of network isolation through DNS-based command-and-control www.beyondtrust.com/blog/entry/p...

GitHub - Richargh/nodatime-mongo-dn-cs-sandbox: Nodatime+Mongo (Sandbox) Nodatime+Mongo (Sandbox). Contribute to Richargh/nodatime-mongo-dn-cs-sandbox development by creating an account on GitHub.

Welcome add-ons but it’s still the broken API behind it. I did a bit of a deep-dive after discovering nodatime: github.com/Richargh/nod...