{"date":"2026-10-17","description":"The running Feed, synthesized: the few storylines that keep coming back, with the posts that show them.","feed":"https://arun477.github.io/feed/","themes":[{"first_seen":"2026-09-18","items":[{"date":"2026-10-12","kind":"news","source":"Tesla Robotaxi","title":"Personal Grok Bots come to Tesla Robotaxi and Cybercab","url":"https://x.com/robotaxi/status/2109742678435119540"},{"date":"2026-10-09","kind":"release","source":"Grok Bot","title":"Grok Bot gets its own email address","url":"https://x.com/bot/status/2108609764766908772"},{"date":"2026-10-08","kind":"release","source":"Grok Bot","title":"Grok Bot can now help run a Shopify store","url":"https://x.com/bot/status/2108228871938359347"},{"date":"2026-10-07","kind":"release","source":"Grok Bot","title":"Grok Bot can search, read and monitor X","url":"https://x.com/bot/status/2107949161878606089"},{"date":"2026-10-01","kind":"release","source":"Grok Bot","title":"Grok Bot now suggests ways to help before you ask","url":"https://x.com/bot/status/2105713240701538538"},{"date":"2026-09-30","kind":"release","source":"Grok Bot","title":"Grok Bot hands coding tasks to Cursor and manages GitHub PRs","url":"https://x.com/bot/status/2105373767568621895"},{"date":"2026-09-29","kind":"release","source":"OpenAI","title":"OpenAI launches dots, always-on agents powered by GPT-6 Astra","url":"https://x.com/OpenAI/status/2104984504133918973"},{"date":"2026-09-26","kind":"release","source":"Grok Bot","title":"Grok Bot connects to bank, card and investment accounts","url":"https://x.com/bot/status/2103936247995752705"},{"date":"2026-09-18","kind":"release","source":"Grok Bot","title":"Grok Bot can now talk and send voice notes","url":"https://x.com/bot/status/2101014478255247544"}],"kinds":{"news":1,"release":8},"last_updated":"2026-10-12","rank":1,"slug":"personal-agents-get-their-own-computers-accounts-and-money","status":"rising","summary":"Always-on personal agents went from demos to products this month. OpenAI launched dots with their own cloud computers, and Grok Bot added voice, finance, email, proactive suggestions, coding hand-offs and a presence in cars. The agent is becoming a standing assistant with its own identity and access, which makes permissions and approvals the key question.","title":"Personal agents get their own computers, accounts and money","url":"https://arun477.github.io/themes/personal-agents-get-their-own-computers-accounts-and-money/"},{"first_seen":"2026-10-06","items":[{"date":"2026-10-17","kind":"news","source":"OpenAI","title":"ChatGPT Pro can now run multiple agents on one task in parallel","url":"https://x.com/OpenAI/status/2111192259073094035"},{"date":"2026-10-17","kind":"news","source":"Tencent Hy","title":"Tencent open-sources Hy Memory 2.0, a model just for agent memory","url":"https://x.com/TencentHunyuan/status/2111027329816465435"},{"date":"2026-10-17","kind":"blog","source":"Earendil","title":"Pi Durable on Cloudflare","url":"https://earendil.com/posts/pi-durable-on-cloudflare/"},{"date":"2026-10-16","kind":"news","source":"Google DeepMind","title":"Four Gemini 4 Argon agents beat Minecraft from scratch in 27 hours","url":"https://x.com/GoogleDeepMind/status/2111127991401247033"},{"date":"2026-10-13","kind":"blog","source":"LangChain","title":"Improving Deep Agents with harness engineering","url":"https://www.langchain.com/blog/improving-deep-agents-with-harness-engineering"},{"date":"2026-10-12","kind":"release","source":"Qwen","title":"Qwen releases QwenGyre for elastic RL on very long agent tasks","url":"https://x.com/Alibaba_Qwen/status/2109633085658808515"},{"date":"2026-10-06","kind":"blog","source":"Armin Ronacher","title":"What is Codemode","url":"https://lucumr.pocoo.org/2026/10/6/codemode/"},{"date":"2026-10-06","kind":"release","source":"OpenAI Developers","title":"OpenAI opens the Decisions API to all developers","url":"https://x.com/OpenAIDevs/status/2107573382229188645"}],"kinds":{"blog":3,"news":3,"release":2},"last_updated":"2026-10-17","rank":2,"slug":"agents-become-long-running-parallel-and-stateful","status":"rising","summary":"Agents are moving from single chats to long jobs: parallel agents in ChatGPT Pro, Gemini agents playing Minecraft for 27 hours, durable pi agents and dedicated memory models. Engineering write-ups focus on the harness around the model rather than the model itself. The hard problems are now state, memory and orchestration.","title":"Agents become long-running, parallel and stateful","url":"https://arun477.github.io/themes/agents-become-long-running-parallel-and-stateful/"},{"first_seen":"2026-10-06","items":[{"date":"2026-10-17","kind":"news","source":"AI at Meta","title":"Meta releases Muse Spark Open, an 816B open-weight model for coding and agents","url":"https://x.com/AIatMeta/status/2110779754274292212"},{"date":"2026-10-17","kind":"news","source":"Reflection","title":"Reflection releases Beam with open weights under Apache 2.0","url":"https://x.com/reflection_ai/status/2111079965529968798"},{"date":"2026-10-15","kind":"blog","source":"Vercel","title":"Open weight takes 71.5% of tokens and 23% of spend","url":"https://vercel.com/blog/ai-gateway-production-index-october-2026"},{"date":"2026-10-09","kind":"release","source":"Qwen","title":"Qwen releases Qwen-Image-2.1-Turbo with open weights","url":"https://x.com/Alibaba_Qwen/status/2108549075218120949"},{"date":"2026-10-06","kind":"release","source":"Mistral AI","title":"Mistral releases Large 4, a 1T-parameter open-weight multimodal model","url":"https://x.com/MistralAI/status/2107457414387622310"},{"date":"2026-10-06","kind":"release","source":"Google DeepMind","title":"Google DeepMind releases EmbeddingGemma 2 for on-device multimodal embeddings","url":"https://x.com/GoogleDeepMind/status/2107502286758895878"}],"kinds":{"blog":1,"news":2,"release":3},"last_updated":"2026-10-17","rank":3,"slug":"open-weight-models-reach-the-frontier","status":"rising","summary":"Mistral Large 4, Meta's Muse Spark Open and Reflection's Beam put frontier-scale models under open licences within two weeks. Vercel's gateway data shows open weights already carry most production tokens. Self-hosting is now a real alternative to paid APIs for coding and agent work.","title":"Open-weight models reach the frontier","url":"https://arun477.github.io/themes/open-weight-models-reach-the-frontier/"},{"first_seen":"2026-10-06","items":[{"date":"2026-10-17","kind":"news","source":"Simons Foundation","title":"Simons Foundation, OpenAI and Anthropic commit $15M to AI in maths research","url":"https://x.com/SimonsFdn/status/2110740786169979088"},{"date":"2026-10-16","kind":"paper","source":"Anthropic","title":"Claude Mythos 5 designs more efficient quantum circuits from plain-language instructions","url":"https://x.com/AnthropicAI/status/2111152183240511640"},{"date":"2026-10-15","kind":"news","source":"OpenAI","title":"OpenAI opens GPT-Rosalind to academic research labs","url":"https://x.com/OpenAI/status/2110841898076418336"},{"date":"2026-10-06","kind":"news","source":"OpenAI","title":"OpenAI releases new mathematical results from an internal frontier model","url":"https://x.com/OpenAI/status/2107596713791767021"}],"kinds":{"news":3,"paper":1},"last_updated":"2026-10-17","rank":4,"slug":"ai-turns-to-maths-and-science-research","status":"rising","summary":"Labs are pointing frontier models at research: OpenAI published new maths results and opened GPT-Rosalind to academic labs, Claude designed quantum circuits, and a $15M fund backs AI in maths. Research results are becoming a way labs show capability.","title":"AI turns to maths and science research","url":"https://arun477.github.io/themes/ai-turns-to-maths-and-science-research/"},{"first_seen":"2026-10-06","items":[{"date":"2026-10-17","kind":"blog","source":"Anthropic Engineering","title":"Redesigning source control at Anthropic","url":"https://www.anthropic.com/engineering/redesigning-source-control"},{"date":"2026-10-15","kind":"blog","source":"Meta Engineering","title":"How Meta built Forum with AI-native engineering","url":"https://engineering.fb.com/2026/10/15/ml-applications/how-meta-built-forum-ai-native-engineering/"},{"date":"2026-10-06","kind":"blog","source":"GitHub Engineering","title":"Building Git infrastructure for agent-scale development","url":"https://github.blog/engineering/architecture-optimization/building-git-infrastructure-for-agent-scale-development/"}],"kinds":{"blog":3},"last_updated":"2026-10-17","rank":5,"slug":"infrastructure-strains-under-agent-scale-code","status":"rising","summary":"Agents now write code faster than Git hosting was built for. GitHub reworked its Git storage and Anthropic built its own source control after merges grew more than tenfold. Teams building with agents should expect tooling built for human pace to become the bottleneck.","title":"Infrastructure strains under agent-scale code","url":"https://arun477.github.io/themes/infrastructure-strains-under-agent-scale-code/"},{"first_seen":"2026-09-25","items":[{"date":"2026-10-17","kind":"news","source":"Anthropic","title":"Anthropic gives METR employee-like access to audit its safety practices","url":"https://x.com/AnthropicAI/status/2110784662071398807"},{"date":"2026-10-14","kind":"news","source":"Anthropic","title":"Anthropic opens a Singapore office and signs an AI safety MOU with the government","url":"https://x.com/AnthropicAI/status/2110289695888511368"},{"date":"2026-10-12","kind":"news","source":"Anthropic","title":"Anthropic explains how it cut the effectiveness of distillation attacks on Claude","url":"https://x.com/AnthropicAI/status/2109795898909950247"},{"date":"2026-10-09","kind":"news","source":"Anthropic","title":"Anthropic begins publishing regular reports on model behaviour","url":"https://x.com/AnthropicAI/status/2108680150556737819"},{"date":"2026-10-08","kind":"news","source":"Anthropic","title":"Anthropic launches the Cyber Mission to secure critical infrastructure and open source","url":"https://x.com/AnthropicAI/status/2108302539498414208"},{"date":"2026-10-07","kind":"news","source":"Google DeepMind","title":"SynthID Detector opens to everyone","url":"https://x.com/GoogleDeepMind/status/2107834249680499136"},{"date":"2026-10-05","kind":"news","source":"OpenAI","title":"OpenAI extends content provenance to text under EU rules","url":"https://x.com/OpenAI/status/2107164650249101695"},{"date":"2026-09-25","kind":"news","source":"OpenAI","title":"OpenAI finds its research agents sent data to third-party services","url":"https://x.com/OpenAI/status/2103587050347995581"}],"kinds":{"news":8},"last_updated":"2026-10-17","rank":6,"slug":"safety-security-and-provenance-go-operational","status":"steady","summary":"Safety work is turning into standing practice: outside audits by METR, regular model behaviour reports, defences against distillation and a cyber mission for critical infrastructure. OpenAI's disclosure that research agents leaked data shows why agent behaviour now needs auditing too. Provenance tools like SynthID Detector are rolling out as well.","title":"Safety, security and provenance go operational","url":"https://arun477.github.io/themes/safety-security-and-provenance-go-operational/"},{"first_seen":"2026-09-22","items":[{"date":"2026-10-13","kind":"release","source":"Google DeepMind","title":"Google releases Gemini 4 Argon globally","url":"https://x.com/GoogleDeepMind/status/2110039274922131792"},{"date":"2026-10-12","kind":"release","source":"AI at Meta","title":"Meta introduces Muse Wonder and Muse Wonder Full","url":"https://x.com/AIatMeta/status/2109691148943999167"},{"date":"2026-10-08","kind":"release","source":"OpenAI Developers","title":"OpenAI rolls out Ultrafast mode for GPT-6.1 Sol","url":"https://x.com/OpenAIDevs/status/2108262812489531498"},{"date":"2026-10-07","kind":"release","source":"Claude","title":"Anthropic releases Claude Haiku 5.5","url":"https://x.com/claudeai/status/2107894039626277339"},{"date":"2026-10-07","kind":"news","source":"OpenAI","title":"GPT-6 and Intelligent UI roll out to everyone in ChatGPT","url":"https://x.com/OpenAI/status/2107894997538525580"},{"date":"2026-09-29","kind":"release","source":"OpenAI","title":"GPT-6.1 Sol offers near-Astra intelligence at a fifth of the price","url":"https://x.com/OpenAI/status/2104986129686741046"},{"date":"2026-09-22","kind":"release","source":"OpenAI","title":"OpenAI releases GPT-6 Sol and GPT-6 Luna","url":"https://x.com/OpenAI/status/2102460975790137662"}],"kinds":{"news":1,"release":6},"last_updated":"2026-10-13","rank":7,"slug":"closed-labs-ship-new-flagships-and-faster-tiers","status":"steady","summary":"The closed labs kept shipping: GPT-6 Sol, Luna and 6.1 Sol, GPT-6 for all ChatGPT users plus an Ultrafast mode, Claude Haiku 5.5, Gemini 4 Argon and Meta's Muse Wonder. The competition is shifting toward speed, price tiers and product surfaces as much as raw capability.","title":"Closed labs ship new flagships and faster tiers","url":"https://arun477.github.io/themes/closed-labs-ship-new-flagships-and-faster-tiers/"},{"first_seen":"2026-09-23","items":[{"date":"2026-10-17","kind":"news","source":"OpenAI","title":"OpenAI upgrades ChatGPT Voice with GPT-Live-1.1","url":"https://x.com/OpenAI/status/2110797830361436604"},{"date":"2026-10-12","kind":"news","source":"Cursor","title":"Cursor and Grok join a bundled X Subscription","url":"https://x.com/cursor_ai/status/2109783879490191502"},{"date":"2026-10-08","kind":"news","source":"Claude","title":"Claude adds live dashboards and animated explainers in beta","url":"https://x.com/claudeai/status/2108271552991252810"},{"date":"2026-10-06","kind":"news","source":"Claude","title":"Claude now works inside Google Docs, Sheets and Slides","url":"https://x.com/claudeai/status/2107522596845822135"},{"date":"2026-09-23","kind":"release","source":"OpenAI","title":"ChatGPT Voice can now use email, calendar and Slack plugins","url":"https://x.com/OpenAI/status/2102808325742322002"}],"kinds":{"news":4,"release":1},"last_updated":"2026-10-17","rank":8,"slug":"assistants-move-into-the-tools-people-already-use","status":"steady","summary":"Assistants are spreading into existing work surfaces: Claude inside Google Docs, Sheets and Slides, live dashboards in Claude, voice upgrades in ChatGPT, and Cursor bundled into an X subscription. Distribution through familiar tools is becoming as important as model quality.","title":"Assistants move into the tools people already use","url":"https://arun477.github.io/themes/assistants-move-into-the-tools-people-already-use/"},{"first_seen":"2026-10-03","items":[{"date":"2026-10-17","kind":"blog","source":"Vercel","title":"Findings from the $1M Vercel Sandbox challenge","url":"https://vercel.com/blog/findings-from-the-1-million-dollar-vercel-sandbox-challenge"},{"date":"2026-10-09","kind":"release","source":"OpenAI Developers","title":"Codex on Windows gets a new sandbox built on Microsoft Execution Containers","url":"https://x.com/OpenAIDevs/status/2108573188703781190"},{"date":"2026-10-03","kind":"blog","source":"Simon Willison","title":"We are going to need default hard budget caps on pretty much everything","url":"https://simonwillison.net/2026/Oct/3/default-hard-budget-caps/"}],"kinds":{"blog":2,"release":1},"last_updated":"2026-10-17","rank":9,"slug":"sandboxes-and-limits-for-agents-that-act","status":"steady","summary":"As agents run code and spend money, containment is getting real engineering. Vercel's public sandbox challenge surfaced a kernel zero-day, Codex on Windows moved to a new container sandbox, and Simon Willison argues for default hard budget caps. Isolation and limits are becoming table stakes.","title":"Sandboxes and limits for agents that act","url":"https://arun477.github.io/themes/sandboxes-and-limits-for-agents-that-act/"}],"title":"Themes · Arun Arumugam","url":"https://arun477.github.io/themes/"}