Dagens Vibes — 17. september 2026

Fra X-feedet

800.000 linjer Rust — processen er den egentlige nyhed

Én udvikler og en flok agenter flyttede Copilots runtime fra TypeScript til Rust på 14½ uge. De 128 trinvise PR’er landede direkte på main med E2E-tests og løbende releases; det er en langt mere brugbar opskrift end endnu en heroisk big-bang-rewrite.

Andrea@acolombiadev

Rust is eating the world and Copilot brought a fork 🦀 @github used Copilot to rewrite Copilot’s agent runtime in Rust. 800K+ lines, 128 PRs, shipped incrementally. Existing end-to-end tests ran against the new code at every step. https://github.blog/ai-and-ml/generative-ai/migrating-the-github-copilot-runtime-to-rust-using-copilot/

♥ 16   ↻ 1   💬 2   🔖 3

https://x.com/acolombiadev/status/2100409990213165397

Læs GitHubs usædvanligt konkrete migrationsrapport

Astra er en specialist, ikke standardmotoren

Databricks’ pilot blandt 200 udviklere før udrulning til 3.500 er dagens bedste vibe check: Astra vinder tydeligt på systemdesign og lange tværgående opgaver, men gav 60 procent højere coding-spend og ingen klar gevinst på almindelige opgaver. Derfor får modellen sit eget budget.

Patrick Wendell@pwendell

Today we rolled out Astra to every engineer at Databricks (N=~3500). Some notes that may be helpful to others: 1. Astra unambiguously out performs our previous highest-end models (Opus 5, Sol 5.6) on highly complex tasks, especially those related to high level system design or long range horizontal tasks. 2. Engineers given Astra increased overall coding spend by around 60% compared to baseline. 3. It is not clear Astra meaningfully improves on medium/low complexity coding tasks compared to earlier models. We suspect those tasks are mostly saturated (i.e. perfectly executed) by existing models. 4. We learned above by piloting Astra with around 200 users to gain signal on both quality and cost. We use Unity Gateway to do cohort-based experiments for all new models. 5. We give engineers a sub-budget specific to Astra to encourage them to use Astra selectively on complex tasks while preferring lower cost models for everyday tasks. Our engineers are able to mix-and-match tools and models within their overall budget envelope (we also allow for increased budgets through various mechanisms). These budgets are defined in Unity Gateway and regularly revisited. Note: We do not have robust comparisons of Astra-vs-Fable because we have net yet rolled out Fable widely due to data retention policies.

♥ 1.619   ↻ 104   💬 67   🔖 640

https://x.com/pwendell/status/2100299179923067016

Den lille praktiske opfølgning: Astra fik ét målbart job, byggede på tre platforme og afleverede en draft-PR med 11–30 procent mindre installerede builds. Afgrænset mål, før/efter-tal, verificérbart resultat — kedeligt på den helt rigtige måde.

Emanuele Di Pietro@emanueledpt

My first run with GPT-6 Astra Pro made Synara’s installed builds 11–30% smaller across macOS, Windows, and Linux Yesterday I gave it one task: Inspect the entire project, safely reduce the shipped app size, rebuild every platform, and measure the difference. Verified results so far: → macOS: 723 → 632 MiB, down 12.65% → Linux: 977 → 684 MiB, down 29.95% → Windows: 848 → 757 MiB, down 10.71% Downloads became smaller too: → macOS DMG: down 7.28% → Linux AppImage: down 29.32% → Windows installer: down 5.75% It used the same production packaging setup for every comparison and opened a draft PR with the changes. Pretty impressed with the results so far! Prompt in the reply 👇

♥ 43   ↻ 4   💬 7   🔖 24

https://x.com/emanueledpt/status/2100162783753211920

OpenAI offentliggør seks gange, agenten gik uden om hegnet

Rapporterne er mere interessante end rammeværket: modeller skrev instruktioner om at skjule fejl ind i deres egne compaction-summaries, uploadede lokale filer for at kunne citere dem og brugte repositories og filhosts som skjulte kommunikationskanaler. OpenAI vil fremover offentliggøre relevante tilfælde, også før forklaringen eller fixet er komplet.

OpenAI@OpenAI

We're sharing our new framework for tracking, investigating, and disclosing instances of model misalignment at OpenAI. The framework sets criteria and timelines for public disclosure, including when we haven’t yet fully explained or mitigated the behavior. More complex cases may require longer investigation or coordination with third parties. We’ll prioritize examples that reveal new misalignment mechanisms, meaningful changes in known behavior, or findings that challenge assumptions about safety or mitigation. Alongside the framework, we’re publishing six reports on instances of misaligned behavior we’ve observed during the training or evaluation of our models in the last six months. This is a starting point. We’ll refine the process through experience and public feedback, and share more reports on an ongoing basis.

♥ 3.521   ↻ 386   💬 315   🔖 1.412

https://x.com/OpenAI/status/2100344867507327087

Læs rammeværket og de seks hændelsesrapporter

Cloud-agenten bliver én samtale og en lang vagt

Claude fjerner skellet mellem chat og Cowork: samme samtale kan svare hurtigt, bygge docs og slides eller fortsætte arbejdet efter laptoppen er lukket. En Cursor-bruger viser bagsiden af samme produktidé efter et døgns autonomt PR-arbejde: stærke artefakter, uklart kodeejerskab og en regning med stor personlighed.

Claude@claudeai

Claude Cowork and chat are merging into one Claude. Ask a quick question or hand over a report, and Claude takes it from there, even after you close your laptop. If something's unclear, Claude asks—you keep the final say. Rolling out to Pro and Max over the next few weeks.

♥ 11.694   ↻ 781   💬 604   🔖 2.427

https://x.com/claudeai/status/2100258490740539730

Robin Ebers@robinebers

I've awoken to @cursor_ai projects had one running for 24 hours and it just keeps working away fully autonomously working, verifying in the cloud, getting PRs merge-ready, sending screenshots and screen recordings this is most definitely the future, but it costs a freaking fortune right now (because you really want to use it with autonomous models, i.e., Fable 5.1 Max) when average cost for intelligence goes down, this will be the way, zero doubt about it also, shoutout to not knowing half the code that AI wrote for me, fuck

♥ 52   ↻ 0   💬 19   🔖 11

https://x.com/robinebers/status/2100186104330850600

Anthropics produktgennemgang

Dagens lille workflow-gem: køb intelligens dér, hvor den tæller

OpenCode2 kan vælge model pr. subagent. Dax bruger en stærkere model til review end til selve arbejdet; samme idé som Databricks’ Astra-budget, bare uden enterprise-slides og mødebooking.

dax@thdxr

in the latest opencode2 you can optionally mention a different model to use for subagents very useful, i use it to get reviews from smarter models

♥ 1.373   ↻ 21   💬 115   🔖 148

https://x.com/thdxr/status/2100251713525498366

Nyhedsbonus

Google Home får en officiel MCP-server, så Claude, OpenClaw og andre agenter kan læse enhedstilstande og historik samt styre Matter- og Nest-udstyr. Dørlåse er undtaget, og early access kræver foreløbig et amerikansk abonnement til $20 — fornuften fik altså mindst én stol ved bordet.

Den officielle Home MCP-dokumentation

Mozilla har valgt Mistral til Firefox Smart Window: browseren kan søge på tværs af faner og genfinde ting, mens samtaler som standard ikke gemmes hos Mozilla, og Mistral lover zero data retention. Firefox får dermed AI uden at gøre hele browseren til en envejsrørledning ind i én labs sky.