AI Pulse: Key AI News — Edition #19 (December 26, 2025)

This edition focuses on agentic AI moving closer to execution, not just reasoning. We cover new open coding models and CLIs designed for autonomous software engineering, IDEs that can visually inspect and modify live UIs, and platforms reframing integrations as first-class apps. We also look at how AI is shifting toward the edge, with function-calling models like Google’s FunctionGemma enabling local automation, alongside faster image generation in ChatGPT and Gemini’s expanding ecosystem for customizable workflows.

🧩 Mistral ships Devstral 2 + Vibe CLI for agentic coding

📅 2025–12–09 🔗 Mistral

Mistral released Devstral 2, a new open coding-model family (including 123B and 24B variants) under permissive licenses, positioned for autonomous, production-grade software engineering workflows. Alongside it, Mistral introduced Vibe CLI, a command-line agent that can navigate, edit, and run changes across an entire repository, bringing agentic coding into a terminal-first workflow.

We’re seeing a clear split between “chat for code” and tool-using agents that operate on whole codebases.

🧱 Cursor adds a visual editor inside the Cursor Browser (v2.2)

📅 2025–12–11 🔗 Cursor

Cursor’s 2.2 release adds a visual editor integrated into the Cursor Browser. It lets you drag-and-drop UI elements, inspect components/props, and apply changes via visual controls, with the agent updating the underlying code. It’s a direct bet that the next IDE loop is: see the running app → edit visually → let the agent reconcile code + structure.

🖼️ OpenAI refreshes ChatGPT Images with GPT-Image-1.5

📅 2025–12–16 🔗 OpenAI

OpenAI launched an upgraded ChatGPT Images experience powered by GPT-Image-1.5, emphasizing faster generation (up to 4x), more precise edits, and improved instruction-following. A dedicated Images space in ChatGPT’s sidebar also ships as part of the update, aiming to make creation and exploration more “productized” (presets, trending prompts, etc.).

🧠 Google introduces FunctionGemma for on-device function calling

📅 2025–12–18 🔗 Google

Google announced FunctionGemma, a specialized variant of its Gemma 3 270M model fine-tuned specifically for function calling. The model is designed to translate natural language instructions into executable, structured actions that can run entirely on-device, without relying on cloud inference.

FunctionGemma supports both conversational chat and deterministic function calls, making it suitable for automating multi-step workflows on resource-constrained hardware while preserving user privacy. Google positions it as a local “control layer” that can orchestrate tools and actions, and optionally hand off more complex reasoning to larger models when needed.

🧩 OpenAI renames “connectors” to “apps” in ChatGPT

📅 2025–12–17 🔗 OpenAI Help Center

OpenAI officially updated its documentation to rename connectors to apps, positioning it as a unified umbrella that covers interactive in-chat experiences and data/tool connections. This change also aligns with the broader push toward an App Directory experience inside ChatGPT, where apps can be browsed/connected based on plan and region.

🧪 Google Labs brings Opal mini-app building into the Gemini web app

📅 2025–12–17 🔗 Google

Google Labs announced that Opal (its no-code, prompt-driven mini-app builder) is now available inside the Gemini web app to create experimental, reusable mini apps.

⚡ Google rolls out Gemini 3 Flash as the fast, efficient option in Gemini

📅 2025–12–17 🔗 Google

Google introduced Gemini 3 Flash in the Gemini app, framing it as a major speed/efficiency upgrade, with “Fast” mode for quick responses and “Thinking” mode for harder problems.

📬 Google Labs launches CC, an email-first productivity agent (waitlist)

📅 2025–12–19 🔗 Google Labs

Google Labs announced CC, an experimental productivity agent that connects to Gmail, Calendar, and Drive to send a personalized daily briefing and help draft emails or create calendar actions via email. The experiment is currently waitlist-based and prioritizes eligible users in the United States and Canada (consumer accounts; 18+).

🔁 Missed **AI Pulse: Key AI News — Edition #18**? That one covered the bigger platform shifts (GPT-5.2, Sora Android shipping fast with Codex, and more) that set the stage for this week’s agents + interfaces acceleration.

If this roundup helps you stay oriented, follow AI Pulse, share it with your team, and comment: are you betting more on visual IDE editing (Cursor-style) or terminal-native agents (Vibe CLI-style) for real production work?