Over 800 subscribers
This Week in All Things AI covers key developments in models, agents, tools, infrastructure, and policy curated from discussions in the All Things AI Telegram group.
If you follow AI for work, research, investing, or just to understand where the technology is heading, this weekly brief is a concise way to scan the most important launches, risks, and resources in a few focused minutes.
Across the week, we saw advances in agentic coding and web tooling (Cursor’s Composer 2.5, Google’s Gemini 3.5 Flash, Modern Web Guidance), major moves in AI infrastructure and pricing (Anthropic’s expanded Colossus 2 deal with SpaceX, DeepSeek’s permanent discount on V4‑Pro), a wave of creative tools (Krea 2 and several AI‑assisted video editors), and notable policy and geo developments from Pope Leo XIV’s forthcoming AI encyclical to Singapore’s latest AI initiatives.
The sections that follow walk through these items day by day, with short context and links so you can dive deeper into the pieces most relevant to your work or interests.
Cursor’s biggest model upgrade yet. Strong gains in long-running agentic coding tasks, reliability with complex instructions, and up to 10x better cost efficiency. It’s built on Moonshot’s Kimi K2.5 and reportedly performs on par with top frontier models in some real-world evaluations. Cursor is also training a much larger model from scratch with SpaceXAI on a massive Colossus 2 cluster.
A substantial improvement in intelligence and behavior over Composer 2, particularly on long-horizon agentic tasks.
The Vatican announced that Pope Leo XIV’s first major encyclical on AI will be released on May 25. It will focus on human dignity and “the protection of the human person in the age of AI.” It is being prepared in collaboration with an Anthropic co-founder.
The pope's encyclical will address 'the protection of the human person in the age of AI', the Vatican says
Pope Leo XIV and the co-founder of artificial intelligence company Anthropic will launch the pontiff's first encyclical on May 25, a document on the care of human dignity in the era of AI, the Vatican said Monday.
Cloudflare tests Mythos against 50+ repositories, highlights its ability to chain bugs into a single exploit, and details a vulnerability discovery harness
In recent weeks, we pointed Mythos and other security-focused LLMs at live code across critical parts of our infrastructure. We share what we observed, the models' strengths and weaknesses, and what the work around them needs to look like before any of it can scale.
Krea 2 (also called K2 or Krea 2 Large) is Krea AI’s first proprietary foundation image model
It was built entirely from scratch with a deliberate focus on aesthetics, style transfer, creative control, and artistic exploration rather than literal prompt transcription or maximum photorealism
It forms part of the broader Krea.ai platform—an all-in-one creative AI suite that aggregates 100+ models for images, video, 3D, editing, upscaling, and fine-tuning under one subscription
They are offering unlimited Krea 2 generations to all their subscribers for the entire week.
Create expressive AI images with Krea 2, Krea AI's in-house foundation model for aesthetic diversity, style control, moodboards, and creative workflows.
via Robby Yung
Gemini Spark is Google's new cloud-based 24/7 personal AI agent announced at I/O 2026, designed to autonomously handle tasks in the background even when devices are off while requiring user approval for major actions.
Initial rollout targets trusted testers this week and U.S. Google AI Ultra subscribers next week in beta, with enterprise availability planned through Google Workspace and Gemini Enterprise.
Summer roadmap includes MCP integrations, macOS support, email/text interaction, Chrome browser control, and subagent creation to expand proactive capabilities.
Google Unveils Gemini 3.5 Flash at I/O Conference
Google launched Gemini 3.5 Flash, generally available now and touted as its strongest model for agentic workflows like multi-step tool use and terminal coding. Benchmarks show it topping prior versions on key tests, holding ground against rivals like Claude 3.5 Sonnet, though independent reviews note mixed results on cost and performance.
The event also introduced Gemini Omni for video creation, Spark as a 24/7 personal agent, developer tools like Antigravity 2.0, plus consumer updates to Search and the Gemini app—ushering in what execs call the agentic era.
At Google I/O we released Gemini 3.5, our latest series of models combining frontier intelligence with action.
Google says its more efficient Gemini 3.5 Flash is the key to your agentic AI future.
Google Launches Modern Web Guidance for AI Coding Agents
At Google I/O 2026, the company unveiled Modern Web Guidance, an early preview toolset that equips AI agents like Claude Code and GitHub Copilot with skills for the latest browser APIs, layouts, and performance techniques. It tackles legacy patterns by injecting concise prompts drawn from Chrome's Baseline initiative, boosting agent success on modern tasks from 52% to 86% in early tests. Backed by Chrome and Microsoft Edge teams, plus experts like Lea Verou, it offers easy installation via a simple npx command and covers real-world uses from CSS animations to WebAuthn logins.
Game changer for Web Development announced at GoogleIO- Modern Web Guidance! It’s expert-vetted skills for web development based on best practices of latest specs and APIs.
Guidance on how to build for the modern web.
Q&A with Harvey CEO Winston Weinberg on launching the legal AI startup in 2022, how AI could shake up law firms' business models, legal AI competition, and more
https://www.ft.com/content/47eafdcb-5a0a-4515-a07a-70e89ad2fa78?syn-25a6b1a6=1
Anthropic is paying SpaceX $1.25B/mo. until May 2029 under their compute deal; Anthropic says it's expanding the deal to include Colossus 2 capacity
FYI, in case some in this group find the below useful for their personal and professional workflows. Personally I consider video-editing as something akin to 'magic of the dark arts' so huge respect to practioners of this craft.
The below are a links to websites of AI assisted video editing tools I've come across (in no particular order), pretty sure these are but a small fraction of tools out there.
Within this list I'm not mentioning the usual suspects such as Premiere, Final Cut Pro, Davinci Resolv, Capcut
Create videos for Socials in 5 minutes. Don't waste time looking up aspect ratios or learning heavy video editors - Keyframes Studio takes care of everything.
Create viral short videos for TikTok, Instagram, and YouTube. Turn ideas into content instantly. Ideate, create, edit, publish - all in one. No skills needed.
Photo and video enhancement software powered by deep learning gets you the best image quality available for noise reduction, sharpening, upscaling, and more.
Meet Capsule-the AI-powered video editor made for content and marketing teams. Create stunning videos 10x faster, collaborate seamlessly, and stay on brand.
Descript makes editing video and audio as easy as editing text. Record, transcribe, edit, and publish in one tool. Try for free, with powerful upgrades for creators & teams.
Use Overlap to create AI clipping workflows to turn long videos into clips, shorts, and social posts automatically. Scale short form content creation with Agentic workflows.
A fast, collaborative, and agentic video editor that runs in your browser. Create, cut, and compose videos with AI - no installs, no downloads.
Eddie is the AI assistant video editor trusted by 40,000+ video professionals. Import interviews and chat with Eddie to get rough cuts and edits in seconds. Then export to MP4, Adobe, FCP, and DaVinci Resolve.
Transform your video editing workflow with Ponder.AI's revolutionary AI-powered platform. Sign up free or download now to experience the future of video creation.
OpusClip turns long videos into high-quality viral clips, and publishes them to all social platforms in one click. We help 10M+ creators create and grow faster.
If any of these do end up being used by you/your team, do share your experiences with this group so that others can also benefit from your learnings
via Alex
SpaceX S1 fillings is a good read for SciFi fans like me. Emphasizing Fiction here 😂
Claiming capabilities to launch 100GW of compute to LEO when hyperscalers are barely able to do 1GW on earth. Claiming 220B daily token usage when the actual xAI usage numbers seem to be more like 15B tokens daily. Claiming the TAM for xAI is over 20T$.
https://www.sec.gov/Archives/edgar/data/1181412/000162828026036936/spaceexplorationtechnologi.htm
via Alba Chung of Alibaba Cloud
Qwen3.7-Max — now available viaAlibaba Cloud Model Studio:
- frontier coding agent: from frontend prototyping to complex software engineering
- office productivity and workflow automation via MCP and multi-agent orchestration
- sustained autonomous execution across long-horizon tasks
-cross-scaffold generalization across diverse agent frameworks
My LinkedIn: https://www.linkedin.com/in/albachung
Feel free to PM me for more details.🙌🏼
Singapore firing on all cylinders
• OpenAI first AI lab outside of the US is Singapore
• Anthropic’s lead investor for their Series G is GIC (Singapore’s sovereign wealth fund)
• NVIDIA chose Singapore for their research hub
• Google Deepmind is strengthening its partnership with Singapore
OpenAI for Singapore launches a multi-year AI partnership to expand deployment, build local talent, and support businesses and public services with AI.
Google DeepMind and Singapore partner to apply frontier AI to address challenges across health, education, sustainability and more through the National Partnerships for AI initiative.
Zero to Claude Code Course Draws 18,000 Beginners Worldwide
The Zero to Claude Code course guides users from terminal basics to advanced AI techniques like prompt engineering and subagents, with hands-on projects including a real-time game capstone. Launched recently at zero2claude.dev it has enrolled over 18,000 students globally, running on a high-performance site that Itay Shmool, VP @ WIX created solo with Claude Code
In the world of increasing prices for models, Deepseek says that they have made their 75% discount originally ending May 31st 2026 for V4-Pro permanent if one uses DeepSeek's inference endpoint.
Recommend group members do their own research to interpret the terms of service/privacy policy under which Deepseek's own inference endpoint is served Deepseek v4-Pro is an open-source/open weight model so there are many inference providers which serve this model and each inference provider serves it under their own ToS/privacy policy
via Robby Yung
an AI demo where Gemini Omni transforms a Google Maps screenshot with a drawn route into a realistic first-person taxi driving video through city streets, complete with accurate buildings, traffic, and dashboard details.
Below is my personal website which aggregates links to many of my socials as well as the various content and community that I curate. Feel free to share this link to others who you think may find this content/community useful to them
The cover image of this newsletter via generated via the OpenAI's ChatGPT Images 2.0 model within the Freepik tool via the following prompt
A fascinating view through a natural rock formation by the sea, which serves as a frame for the scene. The dark, jagged rocks are visible in the foreground, lapped by gentle, turquoise waters. A luxurious yacht can be seen through the opening in the rock face, lying calmly on the clear blue water. Behind it stretches a hilly landscape that glows in soft colors under bright daylight. The contrasting composition between the rough, dark rocks and the elegant, modern yacht creates a harmonious balance of natural beauty and luxurious lifestyle. (Nature and luxury, rich in contrast, maritime scene, turquoise water)
Yusuf Goolamabbas
13.2K
