Dagens Vibes — 6. august 2026

Dagens feed har to store bevægelser: AI-forskning bliver pakket som automatiske eksperimentloops, mens coding agents bliver til vedvarende, auditerbare arbejdsmiljøer. Og rundt om dem vokser det praktiske lag af rettigheder, model-routing og menneskelig review. Chatboksen er ved at blive det mindst interessante sted i bygningen.

Fra X-feedet

Jeff Dean forlader Google efter 27 år og starter Discovery Loop med Sanjay Ghemawat, Oriol Vinyals og Quoc Le. Første mål er at automatisere hele ML-eksperimentloopet — foreslå, køre, måle og gentage tusindvis af forsøg parallelt — før turen går videre til science og engineering. Det er et hold, man ikke bare parkerer i mappen “endnu en AI-startup”.

Jeff Dean
Jeff Dean@JeffDean

Tomorrow will be my last day at Google after 27 years, and watching it grow from 25 people to 190,000+ has been an amazing journey. Below is a note I shared with many people internally at Google today. An excerpt is: It has been an absolute pleasure to work with you and to help build some of the most widely used and impactful products of all time. As a kid, I dreamed of helping build software that would be used by many people, and Google now has thirteen products used by more than a billion people (amazing!). Our work has had a tremendous impact in the world, and I have been lucky enough to collaborate and form friendships with many colleagues that I deeply admire, respect, and enjoy. It still brings me joy every time I see people out in the world using our products to find information, handle email, translate documents, watch videos, learn new things, navigate and understand the physical world, browse the web, use their phone, run large-scale computations on our infrastructure, ride in an autonomous vehicle, or perform complex tasks with the help of our AI systems. I hope you all share this sense of joy, because it is a shared accomplishment! Thank you to all of my colleagues at Google over many years! Now I'm excited to go start @DiscoLoopAI with my longtime friends and colleagues @Sanjay_Ghemawat, @OriolVinyalsML, and @quocleix. (Updated post: slightly redacted to not have some personal info)

Medie fra @JeffDean
♥ 12744↻ 1052💬 407🔖 2328
https://x.com/JeffDean/status/2085083442669318443
Discovery Loop — Continuous ExplorationOfficiel lancering · automatiserede eksperimentloopshttps://www.discoveryloop.com/

Meta lancerer Muse Code og Muse Spark 1.2. Det mest interessante er harnesset: vedvarende baggrundsagenter, et append-only event-log og sessions, der kan replayes og fortsætte præcist efter et crash. En tidlig tester siger dog “harness good, model isn’t great” — ny motor, prøvetur før fanfare.

Mark Zuckerberg
Mark Zuckerberg@finkd

Releasing Muse Code in beta today. It's a terminal coding agent that takes on complete software engineering tasks across large repos: planning changes, writing code, validating the results. Powered by Muse Spark 1.2, a coding-focused model update. https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2

Medie fra @finkd
♥ 11038↻ 952💬 920🔖 1723
https://x.com/finkd/status/2085080750034940201
M
Introducing Muse Code and Muse Spark 1.2Meta AI · 5. august 2026 · terminal-agent og long-horizon modelhttps://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2

Cloudflare OS går efter agentlaget for hele virksomheden, ikke kun udviklere. Agenter starter uden adgang, credentials holdes væk fra genereret kode, og policy følger de data, agenten faktisk har set. Endelig governance, der er mere konkret end “vi har sat MCP foran databasen og krydser fingre”.

rita kozlov 🐀
rita kozlov 🐀@ritakozlov

introducing @cloudflare os, an open source ai productivity platform there are a lot of tools to help developers be more productive with AI, but ramping the rest of the workforce is still a major challenge you need built-in context, skills AND access to systems of record. handing out thousands of API keys is a recipe for disaster this is where what we built is truly unique because it has a built-in governance model that makes it easy to delegate existing permissions and narrowly scope them for the particular task and because it's open-source, you can customize it to fit your orgs needs. the reason everyone seems to be building one of these is: one size fits all doesn't cut it for something that every employee needs to touch

♥ 36↻ 7💬 2🔖 7
https://x.com/ritakozlov/status/2085026288855605315
CF
Cloudflare OS: an open platform for agents, apps, and workCloudflare · 5. august 2026 · open source, Gatekeepers og data-policyhttps://blog.cloudflare.com/cloudflare-os/

Dagens mest direkte workflow-idé: DeepSeek som arbejdshest og Luna som øjne inde i Codex. Det er ikke en benchmarktabel, men model-routing i levende live: forskellige modeller får forskellige roller, fordi harnesset er åbent nok til at blive skruet fra hinanden.

kache
kache@yacineMTB

I really appreciate codex being so open, I was able to add deepseek to it and now I'm editing its source to allow usage of two different providers, so that deepseek can call luna as its eyes I highly recommend trying out codex and maining it!

♥ 1796↻ 46💬 103🔖 493
https://x.com/yacineMTB/status/2085016520841728013

/human-review flytter feedback ud af chatten: åbn HTML eller Markdown lokalt, redigér teksten direkte, resize billeder og kommentér på præcise elementer, før hele pakken sendes tilbage til agenten. Mennesket tilbage i loopet — nu med markør i stedet for metakommunikation om tredje afsnit.

Peter Yang
Peter Yang@petergyang

My last open-source skill, /no-ai-slop, clearly hit a nerve with 4K GitHub stars. Today, I’m introducing /human-review, another free AI skill I think you’ll love. Let’s face it - giving feedback to AI in chat is painful. Asking it to “update the 3rd paragraph” or “update image size to X” is annoying when I just want to edit the text myself or give feedback to AI directly on the page. /human-review opens HTML and Markdown files in a visual editor that lets you: → Edit and format text directly → Resize images → Leave comments for AI like a Google Doc I've been using it to edit PRDs, landing pages, product copy, and much more. The whole review loop runs locally, so your data is safe and nothing is uploaded to the cloud. 📌 Get /human-review for free on GitHub: https://github.com/petergyang/human-review If you find it useful, please ⭐ the repo so more people can find it. 📌 Read my full post here for more on how I use /human-review: https://creatoreconomy.so/p/use-my-human-review-skill-to-edit-html-markdown-visually

Medie fra @petergyang
♥ 935↻ 49💬 29🔖 1699
https://x.com/petergyang/status/2085006701984698712
HR
petergyang/human-reviewGitHub · lokal visuel review af HTML og Markdownhttps://github.com/petergyang/human-review

Matt Pococks skills v1.2 gør skill-økosystemet mindre hjemmelavet: dokumentation, Claude-plugin og Codex-metadata side om side. De små gems er /wizard til menneskekrævende setup-trin og /wait-what til at få modellen tilbage på menneskesprog.

Matt Pocock
Matt Pocock@mattpocockuk

mattpocock/skills v1.2 is out! We're now the 19th most-starred repo of all time. 13.5m downloads on skills​.sh. Thanks for your support! Here's what's new: - Docs: the community's biggest ask. Every skill documented, with explanations of the main flows + troubleshooting - Claude Plugin: install via Claude's official marketplace - Codex Support: full Codex support via agents/openai.yaml files Updated Skills: - /grilling now asks you questions in rounds, not one-by-one - /prototype now uses HTML instead of a TUI for building logic prototypes - easier to share and far richer - /writing-for-agents renamed from /writing-great-skills, use it for ANYTHING your agents read (AGENTS.md, system prompts, docs) New Skills: - /wizard: tired of provisioning infra? Get your agent to build you a TUI to walk you through it - /to-questionnaire: hit a grilling question you can't answer? Turn the session into a doc you can walk through on a call with a colleague - /wait-what: no idea what the model said? Refocus it in your domain language and simplify with ASD-STE100 Full changelog + docs below. Video soon!

Medie fra @mattpocockuk
♥ 4376↻ 317💬 118🔖 3670
https://x.com/mattpocockuk/status/2084985277102031137

Og så den stille perle: Codex fik 200 undervands-RAW-filer og byggede selv en redigeringspipeline — udvalgte billeder, skrev NumPy-scripts, maskerede, vurderede med vision og sorterede arter. Seks likes, langt mere interessant end de fleste virale “lav mig en app”-demoer.

Anshu
Anshu@anshuc

GPT-5.6 Sol is pretty good at… underwater photography? I threw 200+ RAWs I shot this morning into Codex and told it to edit them however it likes, just follow its artistic eye. I didn't expect much, but it did some cool stuff! - Chose 37 with potential - Reframed subjects in interesting ways - Wrote numpy scripts to balance color levels, exposure, black point - Did lots of photo-specific masking and fine-tuning, noise reduction, etc - Reviewed images with native vision and iterated - Classified and organized by species I’m no expert, but these don’t look far off what I’d have done. Curious if you think a pro would’ve interpreted these differently or better?

Medie fra @anshuc
♥ 6↻ 0💬 1🔖 0
https://x.com/anshuc/status/2085002941598232917

Nyhedsbonus

Anthropic bygger et custom-silicon-team for at co-designe chips og Claude. De beholder AWS, Google, Nvidia og AMD i mixet, men signalet er klart: frontier-labs ender som full-stack infrastrukturselskaber, fordi pris og latency også bliver vundet nede i silicium.

AI
Anthropic is hiring an AI chip design teamTechCrunch · 5. august 2026 kl. 16:13 CESThttps://techcrunch.com/2026/08/05/anthropic-is-hiring-an-ai-chip-design-team/