AI Weekly Recap (Week 38)

Plus: The most important news and breakthroughs in AI this week

Author

Alvaro Cintas
September 21, 2025

Good morning, AI enthusiasts. AI just had a wild week. Meta just unveiled smart glasses controlled by the Neural Band, letting you interact without touching a screen. Meanwhile, OpenAI upgraded Codex with GPT-5, making coding smarter and faster than ever.

Plus: The most important news and breakthroughs in AI this week

The new glasses feature a built-in display for apps, directions, and alerts, controlled by the Meta Neural Band, a wristband that senses subtle hand movements using electromyography (EMG). Starting September 30, the glasses will cost $799.

→ Display supports Meta apps like Instagram, WhatsApp, and Facebook
→ Provides directions and live translations directly on the lens
→ Neural Band offers 18 hours of battery life and is water-resistant
→ Uses EMG to detect hand gestures for intuitive control

🧰 Who is this useful for:

  • Early adopters seeking futuristic, hands-free control

  • Content creators wanting higher-quality video capture

  • Athletes tracking real-time performance metrics

  • Anyone looking for stylish AR wearables with long battery life

OpenAI is rolling out GPT-5 Codex across Codex products, GitHub, IDEs, terminals, and ChatGPT, giving developers more powerful tools for writing, debugging, and reviewing code. Unlike earlier versions, GPT-5 Codex can flexibly spend anywhere from a few seconds to seven hours on a task.

→ Dynamically adjusts thinking time
→ Outperforms GPT-5 on coding benchmarks
→ Fewer incorrect code review comments
→ Available now on ChatGPT; API coming soon
→ Competes with Claude Code, Cursor, Copilot

🧰 Who is this useful for:

  • Developers needing faster, more accurate coding help

  • Teams building complex projects requiring long reasoning chains

  • Startups looking to accelerate MVP development

  • Power users who want fewer hallucinations in code reviews

OpenAI and Anthropic just shared new data on how people use ChatGPT and Claude, showing sharp differences in personal vs. work use, global adoption, and task preferences.

→ Claude = coding focus; ChatGPT = writing & decision support
→ Personal ChatGPT use jumped from 53% → 73% (2024–2025)
→ ChatGPT adoption in low/mid-income countries growing 4x faster
→ Claude usage concentrated in wealthy regions
→ Both tools see more task delegation and “information seeking”

🧰 Who is this useful for:

  • Founders exploring global AI adoption opportunities

  • Teams building AI products with region-specific focus

  • Researchers tracking personal vs. enterprise AI usage shifts

  • Creators curious about how AI is shaping daily life worldwide

Google introduced AP2, a new open protocol for AI-driven purchases, backed by 60+ merchants and financial institutions including Mastercard, American Express, and PayPal.

→ Enables AI agents to shop, negotiate, and complete transactions on behalf of users
→ Requires “intent mandate” + “cart mandate” for traceable and auditable purchases
→ Supports complex real-time agent-to-agent negotiations (e.g., travel, bundles)
→ Includes crypto integration via Coinbase, MetaMask, and Ethereum Foundation)

🧰 Who is this useful for:

  • Consumers wanting AI to safely handle shopping or payments

  • Fintech & crypto companies building AI-enabled payment tools

  • Developers creating AI agents that interact with real-world money

  • Businesses exploring autonomous AI workflows for transactions

Alibaba unveiled Tongyi DeepResearch, the first fully open-source web agent achieving performance comparable to OpenAI’s Deep Research using just 30B (Activated 3B) parameters.

→ Scores 32.9 on Humanity’s Last Exam, 45.3 on BrowseComp, 75.0 on xbench-DeepSearch
→ Supports multi-step reasoning via ReAct Mode and complex research tasks via Heavy Mode
→ Real-world applications include Gaode Mate (navigation AI) and Tongyi FaRui (legal research AI)

🧰 Who is this useful for:

  • Researchers and students wanting advanced AI reasoning tools

  • Developers building AI-powered navigation or legal solutions

  • Companies exploring autonomous AI research agents

  • Open-source enthusiasts seeking high-performance AI without massive models

Google is rolling out Gemini in Chrome to all U.S. desktop users, after previously limiting it to premium subscribers. The update adds AI-powered browsing tools that clarify complex info, compare across tabs, and integrate directly with apps like YouTube, Maps, and Calendar.

→ Gemini can summarize and compare info across multiple tabs
→ Will retrieve past browsing sessions on request
→ AI Mode in address bar enables conversational search
→ Built-in protection against scams and fake alerts
→ One-click password resets on supported sites

🧰 Who is this useful for:

  • Everyday users who want simpler, AI-powered browsing

  • Researchers comparing info across multiple sources

  • Shoppers planning purchases across many tabs

  • Anyone needing safer, faster, more intelligent browsing

Notion just introduced AI Agents that automate tasks and analyze data across all your Notion pages and databases, creating reports, meeting notes, and feedback pages.

→ Generates and updates pages, databases, and views automatically
→ Can pull data from Slack, email, and Google Drive for tasks like bug tracking
→ Handles multistep tasks for up to 20 minutes across hundreds of pages
→ Profile pages allow the agent to “remember” key points and follow custom instructions

🧰 Who is this useful for:

  • Teams needing automated meeting notes and competitor analysis

  • Startups tracking projects and data from multiple sources

  • Knowledge workers wanting AI to summarize and act on information

  • Productivity enthusiasts aiming to save time and streamline workflows

Gamma released 3.0, a major update to the AI presentation tool, making slides smarter, faster, and more automated.

→ Gamma Agent: Edit an entire deck with a single prompt, visualize data, or turn screenshots/notes into polished slides
→ Gamma API: Automate deck creation from meeting transcripts or templates using workflows like Zapier
→ Team & Business plans for organizations, plus Ultra plan for power users

🧰 Who is this useful for:

  • Professionals tired of slow, manual slide creation

  • Teams needing automated, on-brand presentations at scale

  • Power users wanting to push visual storytelling further

World Labs launches Marble, a beta platform that generates persistent, explorable 3D worlds from text or image prompts.

→ Create 3D worlds you can explore indefinitely, with consistent geometry and style
→ Export worlds as Gaussian splats for use in games, VR, or web-based 3D projects
→ Supports diverse styles: realistic, cartoon, anime, sci-fi, low-poly, and more

🧰 Who is this useful for:

  • Visual creators and filmmakers wanting coherent 3D sets

  • Game developers and VR storytellers needing immersive environments

  • Artists experimenting with style iteration and large-scale world building

Zoom unveiled AI Companion 3.0 and AI avatars at Zoomtopia 2025, letting users automate meetings and create realistic avatars for online presence.

→ AI avatars mimic your live video for meetings
→ Companion 3.0 takes notes across Zoom, Google Meet, Teams, and in-person meetings
→ Agentic AI can schedule, suggest skipped meetings, and give in-meeting pointers
→ Custom AI agents enable team collaboration, research, and automated tasks

🧰 Who is this useful for:

  • Remote workers who want a polished presence without being on camera

  • Teams using multiple collaboration platforms

  • Businesses needing smarter notes, insights, and automation

  • Users wanting time saved and meetings optimized

That's it! See you tomorrow

- Dr. Alvaro Cintas