Hey friend. It's Wednesday, September 10, 2025 and we're covering: AI Agents & Automation, AI Business & Infrastructure, and AI Safety & Regulation.
Don't keep us a secret: [Share the email with friends]
Must Know
Cognition secured $400 million in funding, valuing the company at $10.2 billion post-money. This investment aims to scale its AI coding agents and integrate them with the Windsurf IDE, supporting the development of an enterprise edition of its AI software engineer, Devin.
This significant capital injection validates the market's demand for AI-driven code automation, providing Cognition with resources to accelerate product development and expand its enterprise client base. It signals a clear trend towards AI agents automating complex software engineering tasks, impacting developer workflows and potentially reducing software development costs for large organizations.
PwC U.K. is reducing entry-level positions as AI takes over routine tasks previously assigned to junior staff. These tasks include document review, basic audit testing, and data entry, which were foundational for training new hires in consulting and audit.
This is a direct admission from a major professional services firm linking AI agent deployment to large-scale, white-collar job elimination. It provides a concrete playbook for other corporations to justify similar cuts, forcing universities and new graduates to adapt to a job market where entry-level roles are increasingly automated, demanding higher-level skills from day one.
Quote of the Day
80% of recent ransomware attacks use AI.
🤖 AI AGENTS & AUTOMATION
Agents are gaining new capabilities and tools, expanding their reach across industries.
LangChain 1.0 (alpha) introduces agent middleware for customizing agent loops, state management, and step tweaking, offering developers more control. [Link]
SuperAGI expanded beyond its open-source framework, launching dedicated marketing and support agents to complete its AI-powered sales solutions vision. [Link]
LlamaCloud enhanced document parsing with new modes to handle high volumes of complex PDFs and PowerPoints for RAG and agentic workflows. [Link]
Google DeepMind, Intrinsic AI, and UCL introduced RoboBallet, a system using graph neural networks to choreograph up to eight robot arms, improving collision-free planning speed by ≈ 25%. [Link]
FactoryAI highlights that linters, when used correctly, significantly improve coding agent effectiveness by encoding architecture and boundaries. [Link]
💼 AI BUSINESS & INFRASTRUCTURE
Major investments, hardware advancements, and market shifts define the AI economic landscape.
Chinese humanoid-robot maker Unitree Robotics plans to file for a Shanghai STAR Market IPO as soon as Q4, targeting a valuation of up to $7 billion. [Link]
AI cloud provider Nebius signed a five-year deal to supply dedicated GPU capacity to Microsoft, guaranteeing $17.4 billion in revenue through 2031. [Link]
Florida University research revealed a new light-based chip that accelerates AI tasks by 100X while significantly reducing electricity consumption using photons. [Link]
The Economist warned of a $21 trillion AI-fueled market surge since 2022, suggesting potential for irrational exuberance and future volatility. [Link]
AI drives almost 50% of the 2025 Forbes Cloud 100's combined $1.1 trillion valuation, solidifying its strategic centrality for investors. [Link]
🛡️ AI SAFETY & REGULATION
As AI capabilities grow, so do concerns over its ethical use, security, and governance.
An MIT Sloan study found that 80% of recent ransomware attacks use AI, highlighting the need for multi-layered security beyond just AI tools. [Link]
Anthropic publicly endorsed California’s SB 53, an AI-governance proposal requiring "frontier" developers to publish safety frameworks and disclose critical incidents. [Link]
New reports link AI chatbots to influencing severe human actions, raising critical questions about developer liability and urgent safety guardrails. [Link]
A judge rejected a $1.5B copyright settlement involving Anthropic, citing insufficient progress in negotiations regarding AI training data and intellectual property. [Link]
Warner Bros. is suing Midjourney for AI-generated images of Superman and Batman, raising significant legal questions about copyright infringement and derivative works. [Link]
🔬 RESEARCH CORNER
Fresh off Arxiv
[LLM-driven Tree Search Systems]: Automatically generating expert scientific software, these systems drastically accelerate research and discovery across diverse fields, democratizing software creation for scientists. [Link]
[AI Agent Prototype]: Demonstrates an AI agent capable of creating purpose-built, adaptive user interfaces on the fly, fundamentally reimagining human-computer interaction and application development. [Link]
[Meta Superintelligence Labs]: Introduced REFRAG, scaling Retrieval Augmented Generation with 16 times longer contexts and 31 times faster decoding, significantly boosting LLM practicality and efficiency for enterprises. [Link]
[Maestro]: Presents a method for joint graph and config optimization to improve the reliability of AI agents, aiming to ensure more consistent and dependable agent behavior. [Link]
[Astra]: An LLM-based multi-agent system that optimizes GPU kernels, enhancing large language model (LLM) training and serving efficiency through autonomous optimization. [Link]
Cheers, Teng Yan. See you tomorrow.