Hey friend. It's Thursday, September 11, 2025 and we're covering: AI Agents & Automation, AI Infrastructure & Hardware, and AI Models & Capabilities.

Don't keep us a secret: Share the email with friends

Must Know

OpenAI signed a $300 billion cloud computing contract with Oracle over five years, significantly expanding its AI infrastructure and boosting Oracle's stock. Concurrently, Broadcom secured a $10 billion surge in ASIC orders, with OpenAI identified as a key customer for custom silicon mass production by 2026.

This unprecedented investment signals OpenAI's aggressive push to scale its compute capabilities, moving beyond reliance on a single cloud provider and potentially developing proprietary hardware. It directly impacts the competitive landscape for cloud providers and chip manufacturers, setting a new benchmark for AI infrastructure spending and accelerating the race for custom AI silicon.

Replit Agent 3 Achieves Human-Level Coding, Claude Gains File Editing Replit's Agent 3 now performs at human-level for many application development tasks, marking a significant leap in AI-assisted coding autonomy. Separately, Claude introduced capabilities to create and edit Excel, Word, PowerPoint, and PDF files directly within its interface.

These advancements directly impact white-collar productivity and software development workflows, enabling AI to handle complex, multi-step tasks autonomously. Corporations can leverage these tools to streamline operations, reduce manual effort, and redefine roles in software engineering and office administration.

Quote of the Day

80% of recent ransomware attacks now use AI, including deepfakes, phishing, malware generation, and AI-driven password cracking.

MIT Research, via X (formerly Twitter)

🤖 AI AGENTS & AUTOMATION

Autonomous agents are rapidly expanding their capabilities and real-world applications.

  • Lindy now integrates with Fireflies, enabling AI agents to autonomously listen to meetings and handle follow-up actions. [Link]

  • FactoryAI's CLI integrates with VSCode/Cursor, offering Opus 4.1 and GPT-5, streamlining AI agent development within IDEs. [Link]

  • LlamaParse now parses speaker notes from PowerPoint files, injecting them as metadata for improved RAG and agentic automation. [Link]

  • brainTerms.ai features multi-agent "experts" with intelligent reasoning to explore and stress-test product and business strategies. [Link]

  • An AI negotiator successfully haggled down a car price by thousands of dollars, showcasing AI agents' potential in negotiation. [Link]

💻 AI INFRASTRUCTURE & HARDWARE

The race for advanced compute and specialized hardware continues to intensify.

  • PsiQuantum secured $1 billion to develop million-qubit, fault-tolerant quantum computers, signaling investor confidence in disruptive compute technologies. [Link]

  • NVIDIA's Blackwell Ultra GPU series shattered MLPerf benchmarks with a 5x throughput increase, solidifying hardware dominance and offering unprecedented inference cost-efficiency. [Link]

  • China unveiled new brain-inspired AI architecture poised to redefine energy efficiency and computational methods, signaling a strategic focus on novel AI development. [Link]

  • Microsoft invested $17.4 billion in a five-year agreement with Nebius to expand AI-ready capacity by 400 MW across Europe. [Link]

  • Arm introduced the Lumex platform, featuring an SME2-enabled C1 CPU and Mali-G1 GPU designed to accelerate AI processing on edge devices. [Link]

🧠 AI MODELS & CAPABILITIES

New model features and competitive advancements are reshaping the AI landscape.

  • Google's Veo 3 AI video model now supports 1080p resolution and vertical 9:16 videos, with price reductions for accessibility. [Link]

  • Google’s Gemini App introduces “Select and Ask” in Gemini Canvas, enabling visual editing of web apps without coding. [Link]

  • ByteDance claims its new Seedream 4.0 image generation model delivers superior aesthetics and alignment compared to Google's Nano Banana. [Link]

  • Microsoft integrates Anthropic's AI models, diversifying its core AI partnerships beyond OpenAI and hedging against single-vendor reliance. [Link]

  • Anthropic's Claude API now integrates web search, enabling the AI to access current information from across the web. [Link]

🔬 RESEARCH CORNER

Fresh off Arxiv

  • [Meta AI]: A paper demonstrates language models can improve through self-play without new data, achieving similar results to data-trained reinforcement learning. [Link]

  • [Arxiv]: EnvX is a framework leveraging Agentic AI to transform GitHub repositories into intelligent, autonomous agents capable of natural language interaction. [Link]

  • [Arxiv]: ChemBOMAS employs an LLM-enhanced multi-agent system to accelerate Bayesian optimization in chemistry, using knowledge-driven and data-driven strategies. [Link]

  • [Arxiv]: Research explores using LLMs to create a team of agents that jointly solve data analysis-based research problems in particle physics, mimicking human researchers. [Link]

  • [Arxiv]: ASearcher enhances search agents through RL, achieving significant gains on web-based benchmarks and supporting tool calls exceeding 40 turns. [Link]

Have a tip or a story we should cover? Send it our way.

Cheers, Teng Yan. See you tomorrow.

Keep Reading

No posts found