💻 OpenAI drops Codex inside Claude Code

Good Morning! OpenAI just made a wildly aggressive ecosystem play by launching a Codex plugin that lives natively inside its biggest rival, Anthropic's Claude Code. Plus I’ll show you how to turn any photo object into a 3D model for free.

Plus, in today’s AI newsletter:

• OpenAI drops Codex inside Claude Code

• Claude Code Gets Full "Computer Use"

• Alibaba’s Qwen 3.5 Omni Can Hear, Watch, and Clone Voices

• How to Turn Any Photo Object into a 3D Model

• 4 new AI tools worth trying

AI NEWS

🤖 OpenAI Drops Codex Inside Claude Code

OpenAI has open-sourced a new plugin that allows developers to run its Codex coding agent directly inside Anthropic's Claude Code CLI. Rather than trying to force developers to switch environments, OpenAI is going right to where the users already are.

• The plugin adds native slash commands (/codex:review, /codex:adversarial-review, /codex:rescue) to delegate tasks and pressure-test code without ever leaving the Claude terminal.

• It leverages existing local Codex CLI authentication, meaning any ChatGPT subscription or OpenAI API key works immediately with no separate billing or setup.

• Claude Code currently dominates developer adoption with a massive $2.5 billion run rate, though Codex has recently grown to 1.6 million weekly active users.

• The launch had awkward timing, coinciding exactly with a disclosed command injection vulnerability in Codex itself from BeyondTrust's Phantom Labs, highlighting the security risks of wiring multiple AI agents together.

By building Codex directly into Claude Code, OpenAI turns a rival platform into a distribution channel. Now, every time a developer asks Claude's AI to double-check its work with OpenAI's model, OpenAI collects the API fee.

ANGENTIC AI

💻 Claude Code Gets Full "Computer Use"

Anthropic has officially integrated its "computer use" capabilities directly into the Claude Code CLI. Instead of just writing and editing text files, Claude can now open apps, click through UIs, type, and take screenshots to visually debug its own work, all without leaving your terminal.

• Available as a research preview for macOS users on Pro or Max plans (

• It can compile native apps (like Swift or Electron), launch them, and click through every screen to run end-to-end UI tests without needing a test harness like Playwright.

• Handles visual debugging by actively resizing windows, finding where UI elements break, taking a screenshot, and patching the underlying code automatically.

• Can drive GUI-only tools that lack APIs, including design software, hardware control panels, and the iOS Simulator.

Coding agents usually hit a wall the moment a task leaves the terminal and enters a desktop app or system UI. By giving Claude Code the ability to see the screen and control the mouse, Anthropic is pushing agentic AI beyond neat tool boundaries and into the messy reality of end-to-end software development. We are rapidly moving from AI that assists with code to AI that operates like a fully autonomous software engineering firm.

AI MODELS

👁️ Alibaba’s Qwen 3.5 Omni Can Hear, Watch, and Clone Voices

Alibaba just dropped Qwen 3.5 Omni, a native "omnimodal" AI model that simultaneously processes text, images, audio, and video in real-time across 36 languages, putting it head-to-head with frontier models from OpenAI and Google.

• Unlike stitched-together multimodal pipelines, Qwen processes audio-visual data natively. In tests, it analyzed a video with sound in one minute—a task that took ChatGPT 5.4 nine minutes using separate transcription and vision tools.

• Features "semantic interruption," allowing the AI to distinguish between background noise (like a cough or saying "uh-huh") and actual user interruptions for natural, real-time dialogue.

• Includes voice cloning capabilities and "Audio-Visual Vibe Coding," where the AI can watch a screen recording of a coding task and write functional code without needing a single text prompt.

• Outperformed Gemini 3.1 Pro on general audio understanding, reasoning, and translation benchmarks, and expanded its speech recognition to cover 113 languages and dialects.

We are moving past the era of text-in, text-out chatbots. By mastering true real-time, omnimodal processing, Alibaba is signaling a massive shift toward fully interactive AI agents that can seamlessly see, hear, and operate inside our actual workflows rather than just alongside them.

HOW TO AI

🎨 How to Turn Any Photo Object into a 3D Model

In this tutorial, you will learn how to extract an element from a real-world photo and generate a high-quality 3D representation using Gemini's image creation capabilities and Copilot 3D Labs.

🧰 Who is This For

• Product designers needing quick 3D mockups from concept photos

• Game developers looking for reference geometry from real objects

• E-commerce sellers wanting to create interactive product views

• Anyone exploring the intersection of AI image generation and 3D modeling

STEP 1: Prepare Gemini for Extraction

Head over to Gemini and select the "Create image" tool. Before doing anything else, make sure to enable "Thinking Mode." This is crucial because it allows the AI to carefully analyze your input image's geometry and textures before attempting the extraction. Upload the real-world photo containing the object you want to isolate.

STEP 2: Run the Isolation Prompt

With your image uploaded, it's time to provide the specific extraction instructions. Use the following prompt, making sure to replace "[element]" with a specific description of your object:

"Generate an image of the [element] in this image. White background, 3/4 view. Make it 100% identical to the original and fill almost the entire white canvas."

Hit generate and watch Gemini isolate your item onto a clean canvas.

STEP 3: Transfer to 3D Labs

Once Gemini produces the isolated image, review it to ensure it is identical to the original object and properly oriented. Download the high-resolution result to your computer. Now, navigate to the Copilot 3D Labs platform, which specializes in converting 2D images into 3D geometry.

STEP 4: Generate in 3D

Upload the clean, isolated image you just downloaded into Copilot 3D Labs. Initiate the 3D generation process. The AI will analyze the shape and textures of the 2D object to extrude it and apply depth, converting it into a manipulatable 3D model that is ready for editing or viewing.

Microsoft rolls out Copilot Cowork to its Frontier early access program and unveils Researcher's Critique and Council tools, which use multiple models together.

Quinnipiac poll: 55% of Americans say AI will do more harm than good in their day-to-day lives, and 65% oppose building data centers in their community.

Meta is testing an Instagram Plus subscription in a few countries, offering features including anonymous Story viewing and extended 48-hour Story durations.

Apple pulls Apple Intelligence in China, after accidentally launching it in the country; there is no imminent launch as Apple has no regulatory approval.

🎨 Phota Studio: AI for editing and generating personalized photos

🌐 MolmoWeb: Ai2’s open-source AI that can browse the web

🔎 Model Council: Perplexity tool to query multiple AI models at once

⚙️ OpenAI Codex: OpenAI’s coding assistant, now with automations and customizable themes

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas