Agent traffic explodes online & Vercel replaces long-lived tokens - AI News (Jun 19, 2026)

Оригинальный автор：The Automated Daily

Первоисточник：https://www.youtube.com/embed/T0UzjIaPjgA

Please support this podcast by checking out our sponsors: - Consensus: AI for Research. Get a free month - https://get.consensus.app/automated_daily - Lindy is your ultimate AI assistant that proactively manages your inbox - https://try.lindy.ai/tad - Discover the Future of AI Audio with ElevenLabs - https://try.elevenlabs.io/tad Support The Automated Daily directly: Buy me a coffee: https://buymeacoffee.com/theautomateddaily Today's topics: Agent traffic explodes online - HUMAN Security reports agentic automation surging and shifting into high-value, post-login flows—keywords: bot traffic, account takeover, scraping, fraud, agentic AI. Vercel replaces long-lived tokens - Vercel Connect swaps stored environment secrets for short-lived, task-scoped credentials via OIDC—keywords: secretless auth, least privilege, token rotation, Slack, GitHub. Vercel’s eve agent framework - Vercel’s open-source eve framework standardizes production agent ops with durable runs, approvals, and tracing—keywords: agent runtime, OpenTelemetry, sandboxing, evals, MCP. LLMs struggle with bug triage - A large study finds bigger context and more “reasoning” don’t reliably solve real vulns end-to-end—keywords: vulnerability triage, context windows, effort tuning, ensemble judging. OpenAI market share dips - Sensor Tower says ChatGPT fell below 50% share while Gemini and Claude gain, signaling a maturing assistant market—keywords: MAU, churn, monetization, ads, competition. ChatGPT adds better scheduling - ChatGPT’s updated Scheduled tasks aims to make recurring automations more dependable across web and mobile—keywords: workflow automation, task management, reliability, productivity. Noam Shazeer joins OpenAI - Top AI researcher Noam Shazeer moving to OpenAI underscores the continuing talent race among major labs—keywords: hiring, research leadership, frontier models, competition. Mistral doubles down open-weights - Mistral previews a new “fat but sparse” open-weight model and emphasizes sovereignty-friendly deployment—keywords: open weights, VPC, auditability, Europe, independence. Small model claims big reasoning - Weibo’s open-source VibeThinker-3B claims near-flagship reasoning on benchmarks, fueling debate over real-world generalization—keywords: 3B model, benchmarks, compression, math, coding. AI policy fight over Anthropic - EFF argues U.S. policy targeted Anthropic with punitive measures after it resisted government demands—keywords: First Amendment, export controls, retaliation, AI regulation, national security. Amazon probes employee activism - Amazon investigating employees after city-council testimony highlights tensions around data centers and worker speech—keywords: activism, data centers, environmental impact, HR, retaliation claims. NVIDIA pushes XR agent plumbing - NVIDIA XR AI offers open-source infrastructure for low-latency multimodal agents in AR/XR settings—keywords: XR, multimodal, GPU backend, MCP, real-time media. AllenAI predicts future 3D motion - AllenAI’s MolmoMotion forecasts object-centric 3D point trajectories, helping robots and video models anticipate motion—keywords: motion forecasting, robotics, trajectories, dataset, benchmark. What model weights really are - A clear analogy-driven explanation of model weights helps demystify how LLMs store capabilities—keywords: parameters, floating point, training, inference, local models. Backlash against forced AI features - An essay argues AI needs a consent-first reset, criticizing non-consensual training and forced feature rollouts—keywords: privacy, opt-out, creators, data centers, environmental concerns. Vibe-coding hype and real risks - A critique of influencer-driven “vibe-coding” warns of fragile apps, security exposure, and unpredictable costs—keywords: no-code promises, token billing, GDPR risk, hype cycle. -Vercel Connect Launches to Replace Long-Lived Agent Tokens with Runtime, Scoped Credentials (https://vercel.com/blog/introducing-vercel-connect) -Morgan Linton Breaks Down What LLM Model Weights Are (https://threadreaderapp.com/thread/2066946225837109735.html) -Critics Compare Replit’s TikTok ‘Vibe-Coding’ Push to MLM-Style Hype (https://www.whatwelo.st/p/generative-ai-is-having-its-herbalife) -Microsoft Azure Releases 'App Modernization Playbook' for Portfolio Prioritization (https://fandf.co/4uokHqN) -Blogger Says AI’s Biggest Problem Is the Industry’s Disregard for Consent (https://www.xodium.net/2026/06/the-ai-hate-progression.html) -AllenAI Open-Sources MolmoMotion for Language-Guided 3D Object Motion Forecasting (https://huggingface.co/blog/allenai/molmomotion) -Testing Whether More LLM Reasoning Improves Security Vulnerability Triage (https://parsiya.net/blog/llm-thonking/) -Vercel Launches eve, an Open-Source Framework for Production-Ready AI Agents (https://vercel.com/blog/introducing-eve) -Amazon Probes Engineers After Seattle Testimony Opposing AI Data Center Expansion (https:...