How LlamaIndex built sandboxed Agents API template

#1 𝕏

Jeff Dean rolled out Gemini 3.5 Flash globally today, unveiling Google’s latest AI model and inviting users to explore its new capabilities in the linked blog post.

Also covered by: @Simon Willison, @Jeff Dean, @Logan Kilpatrick, @Sundar Pichai, @Josh Woodward

#2 𝕏

LlamaIndex 🦙 built a template for Google’s new sandboxed Agents API that auto-clones a Git repo, installs LiteParse CLI and the LlamaParse SDK plus agent skills, and lets agents autonomously parse unstructured documents.

#3 𝕏

NVIDIA AI launched Nemotron-Labs-Diffusion, a family of diffusion-based language models (3B–14B parameters, including vision-language variants) that generate and refine multiple tokens in parallel. This approach delivers faster inference and better GPU utilization.

#4 𝕏

NVIDIA AI released SANA-WM, a 2.6B-parameter open-source world model for controllable 60-second video generation. It uses one image, text prompt, and camera trajectory to deliver precise camera control on a single GPU.

#5 𝕏

Claude launched self-hosted sandboxes (public beta) and MCP tunnels (research preview) in Claude Managed Agents, enabling teams to run agents inside their own security perimeter with controls applied by default.

#6 𝕏

OpenAI introduced Guaranteed Capacity, a new offering that lets customers secure long-term access to OpenAI compute via dedicated infrastructure, partnerships, and capacity-planning investments.

Also covered by: @Sam Altman

#7 📝 OpenAI News

Advancing content provenance for a safer, more transparent AI ecosystem - On May 19, 2026, OpenAI announced it has become a C2PA Conforming Generator Product and is strengthening provenance by adding C2PA-compatible Content Credentials, adopting Google DeepMind’s SynthID invisible watermarking for images generated via ChatGPT, Codex, and the OpenAI API, and previewing a public verification tool to detect those signals. The company says the multi-layered approach—combining metadata and watermarking—addresses metadata loss (noting prior visible watermarks in Sora and audio watermarking in Voice Engine), the verification tool will initially only cover OpenAI-generated content, and it will avoid definitive conclusions when no provenance signals are detected.

#8 📝 Anthropic News

KPMG integrates Claude across its core business and workforce of more than 276,000 in strategic alliance - KPMG has formed a global alliance with Anthropic to embed Claude into its Digital Gateway platform and give all 276,000+ employees access, using Claude for tax, legal, cybersecurity and private equity work while being named a preferred Anthropic partner for PE. With Claude Cowork, Managed Agents and offerings like KPMG Blaze (embedding Claude Code), KPMG says teams can build AI agents and modernize systems in minutes rather than weeks under its Trusted AI framework, supported by joint research with UT Austin on human-in-the-loop deployment.

#9 𝕏

Google Research unveiled Gemini for Science at Google I/O 2026, a suite of AI-driven tools and experiments designed to dramatically scale up and sharpen the precision of scientific exploration.

#10 in

Dharmesh Shah announces Andrej Karpathy has joined Anthropic to use Claude to accelerate AI research, underscoring the huge leverage of AI-powered research loops.

#11 𝕏

Google DeepMind launched the Build with Gemini XPRIZE Hackathon, offering a $2 M prize pool for developers to build real-world apps that tackle global challenges—from reducing food waste to advancing scientific research.

#12 𝕏

Garry Tan released GBrain—an MIT-licensed OSS retrieval and memory system that achieves SOTA on LongMemEval, beating all known open-source repos by over 1% without any LLM query rewriting. He’s already running it on his own 100k-page OpenClaw/Hermes Agent brain.

#13 𝕏

Julien Chaumond – Co-founder and CTO @HuggingFace lays out how to run Qwen3.6 models on llama.cpp with Multi-token prediction (MTP). By toggling flags like --mtp 1 and --mtp-context 32, he achieves a 2× speedup in inference.

#14 𝕏

Santiago shows how integrating one library—and just one line of code—unlocks full observability and automatic analytics for LLM-based apps, delivering extensive telemetry for free.

#15 📝 PromptLayer Blog

MCP vs API: Architecture patterns for AI agents and applications - Explains the two core protocols—MCPs and APIs—that power AI workflows, and how they differ in enabling agent actions, data lookups, prompt evaluation, and orchestration in production AI systems.

#16 📝 Claude Code Blog

Using Claude Code: The unreasonable effectiveness of HTML - Presents techniques for using HTML with Claude Code to produce structured, reliable outputs and streamline workflows. The article demonstrates practical uses of HTML within Claude Code to improve clarity and reproducibility.

#17 𝕏

Logan Kilpatrick launched major updates to Google AI Studio and the Gemini API—Gemini 3.5 Flash, managed agents with the antigravity harness, native Android app creation in AI Studio, workspace integrations, and one-click antigravity export.

Also covered by: @Simon Willison, @Jeff Dean, @Logan Kilpatrick, @Sundar Pichai, @Josh Woodward

#18 𝕏

Google AI launched a Gemini 3.5-powered intelligent Search box that delivers AI Overviews for concise summaries and an interactive AI Mode for step-by-step, deeper query exploration.

Also covered by: @Logan Kilpatrick

#19 𝕏

Demis Hassabis unveiled Gemini Omni, a multimodal AI that ingests photos, video, and audio to generate and iteratively edit entirely new scenes. It starts with video support and will soon handle any input/output format.

#20 𝕏

Guillermo Rauch shipped a CDN pricing model that “smooths over” traffic spikes and viral events, delivering predictable costs without slow routes, degraded PoPs, or priority bandwidth tiers—all while maintaining uncompromising network performance.

#21 𝕏

Cursor is now integrated into Jira—assign it to work items or mention @Cursor in comments to launch a cloud agent that uses issue titles, descriptions, comments, and your repo settings to auto-generate merge-ready PRs.

#22 𝕏

xAI launched Grok and X Premium support in OpenClaw, letting users chat with their AI agent, generate images and videos, or search X posts directly.

#23 𝕏

There's An AI For That Unframe signed $100M in customer contracts and raised $100M in funding in under a year, making it one of the fastest-growing enterprise AI companies on record.

#24 𝕏

Marily Nika unveiled Google I/O 2026’s agentic Gemini era, rolling out proactive AI agents that perform tasks under user direction. She also showcased the biggest upgrade to Google’s Search Box in 25 years, plus new Omni capabilities and extensive developer resources.

Also covered by: @Logan Kilpatrick

#25 𝕏

Teresa Torres spent 60% of her time building AI-powered discovery tools—partnering with Vistaly to license them and quietly launching “Teresa Bot,” an AI coach trained on her writing.