AI News: A Huge Week for AI Apps (Anthropic, OpenAI, Google)

**TL;DR:** OpenAI’s new Codex desktop app quietly became the closest thing to a “super app” for coding and agentic work, while Anthropic upgraded Claude Code with parallel sessions and Google rolled out desktop apps, Chrome slash commands, and a hyper-expressive TTS model. ## OpenAI Codex Desktop: the stealth “super app” OpenAI released a redesigned **Codex desktop client** (v 5.4) that packages code generation, image creation, browser automation and long-running agents inside a single IDE. ### Key capabilities - **Background computer-use**: multiple agents can click and type on macOS while you keep working in other apps - **Built-in browser**: circle any part of a live preview, type a natural-language note and Codex edits the code immediately - **GPT-image-1.5** is baked in—no need to jump back to ChatGPT - **Session memory**: re-use previous actions and persistent project context ### Live demo 1. “Draw a mock-up for a site that sells surfboards and tacos” → 30-second mock-up generated inside Codex 2. “Code the page from that mock-up” → working site served in the internal browser after 6 minutes (SVG placeholders included) 3. Select hero section → “Generate a background image with a surfboard and food-truck, keep text readable” → new asset swapped in < 2 minutes 4. “Build a local Connect-4 desktop app” → 13 minutes later a runnable .app with difficulty selector and built-in AI opponent 5. “Play two games and QA the experience for me” → Codex controls the mouse, plays itself, writes a summary Take-away: Codex now feels like a full “agent IDE” rather than a side-project. ## Anthropic refreshes Claude Code Anthropic shipped a visual overhaul and new tooling for **Claude Code**: - **Parallel sessions**: pin multiple repos or tasks as tabs and hop between them while Claude keeps working - **Top-left switcher** unifies Code, Co-work and regular chat modes - **Integrated terminal, manual file editor, faster diff viewer** - **Preview pane** renders HTML, PDFs and local dev servers without leaving the app Goal: keep developers inside Claude Code instead of the terminal. ## Google’s triple drop ### 1. Stand-alone desktop apps - **Windows**: new “Google” app bundles Search + AI Mode (rolling out globally) - **Mac**: Gemini app already live; same feature set as the web client—chat, Imagen, VO video, Canvas, Deep Research, guided learning, model toggle (Fast / Pro) ### 2. Chrome “Skills” - Save frequent prompts as slash commands (e.g., `/break this news into bullet points`) and invoke them instantly on any page - Comparable to Perplexity’s Comet feature; staged rollout, not yet universal ### 3. Gemini 3.1 Flash TTS - New speech model available in Vertex AI, Google Vids and AI Studio - Emotional tags—`<excited>`, `<whisper>`, `<panic>`, `<laugh>`, `<dramatic-pause>`—produce natural, controllable voice output - Test it in AI Studio under Audio → “Gemini 3.1 Flash TTS Preview” ## Quick mentions - **Perplexity** teased a “Personal Computer” feature—details still emerging Next week is already shaping up to be even busier; stay tuned. **Source:** [Matt Wolfe on YouTube](https://www.youtube.com/watch?v=bIrzOQtnp8w)