Microsoft just made a ton of new AI announcements across GitHub, Copilot, Azure AI Foundry, Windows, and more.
Here’s everything important announced live from Microsoft Build 2025:
1. GitHub Copilot is going from an in-editor assistant to a fully autonomous coding agent!
It works asynchronously to add features, fix bugs, extend tests, refactor code, and improve documentation
Plus, Microsoft is open-sourcing Copilot Chat in VS Code
2. Copilot Tuning: A new, low-code capability in Copilot Studio to train models and create agents using company data
3. All agents built can now be integrated with Teams and Copilot
Users can chat with them, assign action items, and kick off new workflows by mentioning them in a chat or meeting
Plus, the enhanced Teams AI library now supports MCP and A2A protocols.
4. Azure AI Foundry updated with new models including Grok 3, Flux Pro 1.1, and Sora (coming soon)
Also includes new model router, multi-agent workflows, observability features, and Foundry local for creating localized AI apps on Windows and Mac
5. Windows enhanced with new AI-focused capabilities, including:
—Windows AI Foundry with Windows ML, Foundry Local, ready-to-use AI APIs for vision and language tasks
—Native MCP support
—App Actions
—Open-sourced Windows Subsystem for Linux
6. NLWeb: A new open protocol to make any website an agentic application, capable of supporting AI search
This will allow users to query the contents of the site by using natural language, just like with an AI assistant or Copilot
7. Microsoft Discovery: A new agentic platform that enables researchers to collaborate with specialized AI agents to drive scientific research and outcomes
The agents generate ideas, simulate results, and learn over time
• • •
Missing some Tweet in this thread? You can try to
force a refresh
I’ve been testing it all week, and it’s starting to actually *stick*
Having in-page data context and plugging it directly into tabs makes it so much easier to add to my workflow
An honest review with a few use cases (and caveats):
First, the caveats:
- It's still early, and struggles with longer workflows
- Most data is retained by Perplexity (privacy concern)
- Only available to Max users ($200/mo) or free with an invitation code (we just dropped 500 codes in The Rundown University)
USE CASE 1: Email and Calendar management
With the Google Cal and Gmail integrations, it can summarize emails, schedule meetings, and manager calendar events directly from the browser
AI NEWS: Meta just teamed up with Oakley to expand its family of AI smart glasses.
Plus, more news from Mistral, Moonshot AI, MiniMax, Anthropic, and xAI.
Here's everything you need to know:
Meta launched new performance-focused AI smart glasses with Oakley. Key features include:
—Upgraded camera with up to 3K videos
—2x battery, up to 8 hours of usage & 19 hrs on standby
—PRIZM Lens tech to amplify contrast across changing light and weather
Mistral released Mistral Small 3.2-24B, a minor update for the Small-3.1 model, with:
—Enhanced instruction following capabilities
—More robust function calling
—Fewer infinite generations or repetitive answers
AI NEWS: Sam Altman just revealed Meta is offering "$100M signing bonuses" to poach talent from OpenAI.
Plus, more news from Google, Baidu, MiniMax, Proactor, Krea, and Intelligent Internet.
Here's everything you need to know:
Sam Altman cooked Meta on his brother's podcast. Key takeaways:
— Meta’s offering $100K bonus to poach OAI talent
— None of OAI’s best have taken the offers
— OAI has a better shot at AGI, will eventually be more valuable
— Meta isn't great at innovation
Google just launched Gemini 2.5 Pro and Flash models in general availability
The co also started rolling out a hyper-efficient Flash-Lite model (in preview)
All three feature adjustable "thinking" with Flash-Lite defaulting to thinking off for max speed
AI NEWS: China's Tencent just dropped an AI that generates 3D assets with cinematic quality.
Plus, more news from OpenAI, ElevenLabs, Flowstep, The Alan Turing Institute, and Beijing Academy of Artificial Intelligence.
Here's everything you need to know:
Tencent released Hunyuan 3D 2.1, an open-source model for generating 3D assets, including PBR materials
It can synthesize objects with cinematic textures and realism while covering their light interactions
Fully open-source with model weights and code
OpenAI dropped a few notable ChatGPT updates, including:
—Downloads for Canvas docs and code, including as PDF, docx, or markdown
—A new Projects experience with deep research, voice mode, file uploads via mobile, and improved memory to reference chats
AI NEWS: Apple kicked off WWDC 2025 with a major UI revamp — but was pretty light on Apple Intelligence
Plus, more news from Google DeepMind, OpenAI, Microsoft, Manus, and Skywork AI.
Here's everything you need to know:
At WWDC 2025, Apple showed off only a handful of AI upgrades, including:
—New Live translation for FaceTime, Messages, and calls
—Visual intelligence via screenshots
—AI-powered intelligent actions in Shortcuts
—AI "Workout Buddy" on Apple Watch
Google DeepMind and the UK govt just dropped "Extract," a Gemini-powered tool to expedite infrastructure and housing decisions
It uses multimodal reasoning to turn complex planning docs – including handwritten notes – into digital data in just 40 seconds