Between December 8–13, 2025, many important AI updates were released for developers. I collected the most important updates you should know about. These updates include new AI models, better tools for building AI agents, design tools, and developer platforms.
1. Google Gemini Deep Research & Interactions API
Google released Gemini Deep Research, a new AI agent powered by Gemini 3 Pro. It can read and understand large amounts of information, then create clear reports with sources and citations.
Developers can use this agent through the new Interactions API. This API is a single REST interface that helps manage conversations, memory, and tool usage when building AI agents.
Read more: [https://blog.google](https://blog.google/technology/developers/deep-research-agent-gemini-ap…
Between December 8–13, 2025, many important AI updates were released for developers. I collected the most important updates you should know about. These updates include new AI models, better tools for building AI agents, design tools, and developer platforms.
1. Google Gemini Deep Research & Interactions API
Google released Gemini Deep Research, a new AI agent powered by Gemini 3 Pro. It can read and understand large amounts of information, then create clear reports with sources and citations.
Developers can use this agent through the new Interactions API. This API is a single REST interface that helps manage conversations, memory, and tool usage when building AI agents.
Read more: https://blog.google
2. OpenAI GPT-5.2 Release
OpenAI launched GPT-5.2, calling it their most powerful model so far. It is better at:
- Writing and understanding code
- Handling long conversations and documents
- Working with images and other inputs
- Using tools correctly
There are three versions:
- Instant — fast and lightweight
- Thinking — best for coding and planning
- Pro — highest accuracy for complex tasks
I haven’t tested it yet, but I definitely will.
Read more: https://openai.com
3. Microsoft Ignite Announcements
At Microsoft Ignite, Microsoft introduced Foundry, a platform for building and running AI agents at scale.
Key parts include:
- Foundry IQ for searching and using knowledge
- Foundry Control Plane for security and compliance
- Agent Framework (combines Semantic Kernel and AutoGen)
- Hosted Agents for running agents in production
- Model routing, including support for Anthropic’s Claude
This platform helps teams build, manage, and control AI agents more easily.
Read more: https://techcommunity.microsoft.com
4. Agentic AI Foundation (AAIF)
OpenAI, Anthropic, and Block created the Agentic AI Foundation (AAIF). Its goal is to support open standards and neutral governance for AI agents.
Companies like Google and Microsoft are contributing important technologies, such as:
- Model Context Protocol (MCP)
- AGENTS.md, a standard for describing agent behavior
This helps different tools and platforms work together. If you want to know more about AGENTS.md please watch my explanation video:
Read more: https://openai.com
5. Google Disco & GenTabs
Google Labs introduced Disco, an experimental AI tool. One feature, GenTabs, can turn your open browser tabs and chat history into simple web apps.
Using natural language, GenTabs can create:
- Dashboards
- Small tools
- Interactive pages
No coding is required, and it is powered by Gemini 3.
Read more: https://blog.google
6. Google Stitch with Gemini 3
Google updated Stitch, a design tool now powered by Gemini 3. It helps generate high-quality user interfaces. A new Prototypes feature lets developers connect screens into interactive flows without writing code. This is useful for quickly testing ideas.
If you want to see a video about Stitch, please let me know.
7. Mistral Devstral 2 & Vibe CLI
Mistral released Devstral 2, an open-source coding model with:
- 123 billion parameters
- 256k token context window
It performs very well on coding benchmarks. Mistral also released Vibe CLI, a command-line tool that understands your project and lets you explore and modify code using natural language.
Read more: https://mistral.ai
8. Google CLI Session Management
Google updated the Gemini CLI to automatically save:
- Prompts
- Responses
- Tool calls
- Token usage
These are stored in project-based sessions. Developers can browse and resume sessions using a session browser. This makes long-running or repeated workflows much easier.
This feature is enabled by default in version 0.20.0. Also make sure you watch my video about 10 Gemini CLI Pro Tips Every Coder Should Know!
9. Microsoft Research Agent Lightning
Microsoft Research introduced Agent Lightning, an open-source framework for training AI agents using reinforcement learning.
It separates:
- Agent execution
- Agent training
This means developers can add reinforcement learning without rewriting their code. The system tracks agent actions and assigns rewards for each AI call.
Read more: https://microsoft.com
Conclusion
December 2025 showed how fast AI tools for developers are improving. New models, open standards, and agent frameworks are making it easier to build smarter AI systems with better reasoning, smoother interfaces, and more control.
See you next week! ;)