AI Weekly Recap (Week 3)

Happy Sunday! We just had another crazy week in AI. China just dropped an insane AI image model that’s free and open-source while this new tool lets you clone any website design in seconds.

And that's not all, here are the most important AI moves you need to know this week.

1. MagicPath Launches One-Click Web-to-React Capture

MagicPath has released a Chrome extension that lets developers clone any website element and instantly convert it into an editable React component with a single click.

• Capture any HTML element from the web directly into an editable canvas

• Automatically converts captured HTML into interactive React components

• Edit components manually or modify them using natural-language AI instructions

• Reuse existing project HTML instead of rebuilding components from scratch

• Cut hours of frontend development time by starting from real, production-grade designs

Try it Now → https://chromewebstore.google.com/detail/web-capture-html-to-react/ejendfoehmomnkeedkpmkkcmcalklpcc

2. Manus Releases “Meeting Minutes”

Manus has rolled out Meeting Minutes, a new feature that turns any conversation into structured, usable output. With a single speech-bubble click, you can now record meetings up to 2 hours long, automatically transcribe them, and instantly extract notes, action items, and workflows.

• One-click recording with support for meetings up to 2 hours

• Automatic transcription and translation of the full conversation

• Instantly extract notes, action items, processes, and decisions from transcripts

• Organize all recordings inside a single Manus Project for long-term memory and retrieval

• Turn conversations into task lists, PRDs, strategic plans, or customer personas

Try it Now → https://manus.im/

3. Google Launches Personal Intelligence in the Gemini App

Google has introduced Personal Intelligence, a new feature inside the Gemini app that connects data across Google services like Gmail and Google Photos to deliver deeply personalized answers.

• Connects data across Google apps (Gmail, Photos, and more) to understand personal context

• Reasons across your data to surface proactive insights, not just retrieve information

• Understands nuance without users explicitly telling Gemini where to look

• Available in beta for personal accounts, with rollout to AI Mode in Search later this year

• Initially limited to Google AI Pro and AI Ultra subscribers in the U.S.

Try it Now → https://gemini.google.com/

4. Z.ai Open-Sources GLM-Image

Z.ai has released GLM-Image, a 16B-parameter open-source image generation model designed for text-heavy, information-dense visuals like infographics, slides, and technical diagrams.

• Beats Nano Banana Pro on complex text rendering accuracy (0.91 vs 0.77 on CVTG-2k)

• Maintains >90% accuracy even with multiple text regions in a single image

• Hybrid AR + diffusion design separates layout reasoning from visual rendering

• Excels at infographics, posters, diagrams, and instructional visuals

• Slightly weaker in aesthetics and instruction-following compared to Google’s model

• Fully open-source with MIT / Apache 2.0–friendly licensing for enterprise use

Try it Now → https://chat.z.ai/

5. ElevenLabs Releases Scribe v2

ElevenLabs has introduced Scribe v2, its most accurate transcription model to date, setting a new state of the art in speech-to-text. While Scribe v2 Realtime is optimized for ultra-low latency and agent workflows, Scribe v2 is built for high-accuracy batch transcription, subtitles, and captions at scale.

• Lowest word error rate (WER) on industry-standard benchmarks

• Designed for batch transcription, subtitling, and large-scale captioning

• Handles pauses, long silences, tone changes, and delivery shifts with high stability

• Major accuracy and robustness upgrade over Scribe v1

• Now live inside ElevenLabs Studio for production-grade workflows

Try it Now → https://elevenlabs.io/speech-to-text

6. Anthropic drops Claude Cowork

Anthropic introduced Cowork, a new tool inside the Claude Desktop app that lets users give Claude controlled access to folders, no coding, terminals, or setup required.

• Users assign a sandboxed folder Claude can read and edit via chat

• Built on the same Claude Agent SDK that powers Claude Code

• Designed for non-technical tasks like expense reports, media organization, and analysis

• Ships with safety warnings around prompt clarity, file deletion, and agent autonomy

Try it Now → https://claude.ai/new