Hey friend. It's Wednesday, September 17, 2025 and we're covering: AI Infrastructure & Compute, AI Agents & Automation, and AI in Business & Society.
Don't keep us a secret: Share the email with friends
Must Know
UAE Secures AI Chips Via Trump Crypto Firm Investment A New York Times report reveals the UAE received critical AI chips after depositing $2 billion into a Trump-affiliated crypto firm. This transaction highlights AI chips as strategic national assets. The deal underscores the growing intersection of geopolitics, finance, and access to advanced technology.
This transaction sets a precedent for how nations acquire strategic AI hardware, bypassing traditional supply chains through financial leverage. It signals a new era where access to compute power becomes a key geopolitical bargaining chip, influencing international relations and national security postures. Corporations must navigate this complex landscape to secure their own AI infrastructure.
OpenAI Developers Shift to Voice Commands for AI Agents OpenAI developers report they no longer write code directly, instead using voice commands to instruct AI agents. This marks a significant shift in software development paradigms. The reliance on AI agents for coding tasks accelerates AI's productivity impact across the industry.
This development provides a concrete example of AI agents automating core white-collar tasks, directly impacting software engineering roles. It suggests a future where human-AI collaboration shifts from coding to high-level instruction and oversight. Companies must prepare for evolving workforce needs and invest in training for agentic workflow management.
Quote of the Day
OpenAI insiders claim they 'don’t program anymore,' instead relying on voice commands to AI agents.
🖥️ AI Infrastructure & Compute
The race for AI compute intensifies, driving massive data center investments and new hardware solutions.
xAI plans the world's first gigawatt datacenter, Colossus 2, for its Grok models, underscoring escalating compute requirements. [Link]
GPT-5-Codex demand exceeds forecasts, causing 2x slower speeds and highlighting GPU infrastructure bottlenecks. [Link]
OpenAI, NVIDIA, and Nscale launch Stargate UK, a sovereign AI infrastructure partnership with up to 50,000 GPUs. [Link]
Autodesk Research leverages NVIDIA GH200 to accelerate computational fluid dynamics, improving engineering simulations. [Link]
NVIDIA Run:AI Model Streamer reduces cold start latency for LLM inference, optimizing GPU memory loading. [Link]
🤖 AI Agents & Automation
AI agents are transforming workflows, from coding to commerce, with new protocols and specialized tools.
Google introduces the open-source Agent Payments Protocol (AP2), backed by 60+ partners, for AI agent commerce. [Link]
LangChain releases Summarization Middleware in v1 alpha to prevent context window overflow in long agent loops. [Link]
Workday acquires AI agent company Sana for $1.1 billion, validating specialized AI agent solutions for enterprise workflows. [Link]
Orchids is introduced as the world's first AI Full Stack Engineer, setting state-of-the-art in UI and UX. [Link]
Paper2Agent automatically transforms research papers into active AI agents, streamlining research workflows. [Link]
🏢 AI in Business & Society
AI's societal integration deepens, raising questions about job displacement, ethics, and human-AI interaction.
Fiverr will cut 30% of its staff to become an "AI-first" marketplace, reinvesting in AI for operations. [Link]
Google lays off over 200 AI raters, indicating AI systems are automating human training tasks. [Link]
Reuters and Harvard researchers demonstrate generative AI chatbots can be weaponized for large-scale phishing. [Link]
A MIT + Harvard study indicates human-AI companionship is stable and serious, with positive user experiences. [Link]
Millions are seeking spiritual guidance and confession from AI chatbots, signaling new AI societal integration. [Link]
🔬 Research Corner
Fresh off Arxiv
[University of California, Berkeley]: Causal-Symbolic Meta-Learning (CSML) infers latent causal structures, enabling rapid adaptation to novel tasks from few examples. [Link]
[Tsinghua University]: LTA-Thinker, a Latent Thought-Augmented Training Framework, improves distributional variance and enhances LLM reasoning performance. [Link]
[Google Research]: Graph Reasoning via Retrieval Augmented Framework (GRRAF) uses RAG and LLM code generation for graph reasoning tasks, achieving 100% accuracy. [Link]
[Shanghai Jiao Tong University]: ReTrack introduces a fast data unlearning method for diffusion models by redirecting denoising trajectories, preserving generative quality. [Link]
[University of Cambridge]: A longitudinal study evaluates LLM reasoning, showing performance improvement from hidden Chain of Thought prompting and thinking models. [Link]
Have a tip or a story we should cover? Send it our way.
Cheers, Teng Yan. See you tomorrow.