Hey friend. It's Tuesday, September 16, 2025 and we're covering: Agentic AI & Robotics, AI Business & Policy, and Cutting-Edge Research.
Don't keep us a secret: Share the email with friends
Must Know
OpenAI Launches GPT-5-Codex for Agentic Development OpenAI released GPT-5-Codex, a specialized version of GPT-5 optimized for agentic coding. Available across CLI, IDE extensions, web, and mobile, it already accounts for 40% of Codex traffic. The model offers 10x faster processing for simple queries and allocates double the time for complex tasks, enhancing efficiency in software development and cybersecurity.
This launch provides a significant productivity boost for engineering teams, accelerating code creation and refactoring. Its rapid adoption indicates a fundamental shift in software development workflows, where AI agents handle increasingly complex tasks, reducing development cycles and time-to-market for new products.
Fiverr Cuts Workforce, Shifts to AI-First Model Fiverr reduces its workforce by 30%, eliminating 250 roles, as it transitions to an "AI-first" marketplace. The company plans to automate operations using internal AI systems to increase efficiency and speed. This strategic pivot aims to streamline processes and adapt to the evolving demands of the freelance economy.
This move provides a clear example of AI's direct impact on white-collar employment and corporate restructuring. Other service-oriented businesses will observe Fiverr's success in automating core functions, accelerating similar workforce adjustments across industries to enhance operational leverage and reduce labor costs.
Quote of the Day
GPT-5-Codex is rapidly gaining traction, already accounting for approximately 40% of Codex traffic and expected to become the majority soon.
🤖 Agentic AI & Robotics
AI agents and embodied systems rapidly advance, automating complex tasks.
Meta achieved a 25x speedup in AI agent training, advancing robotics and complex planning applications. [Link]
OpenAI job postings seek expertise in humanoid control and high-volume hardware production, indicating ambitious robotics plans. [Link]
Boston Dynamics' Atlas robot, trained with Large Behaviour Models, demonstrates unprecedented dexterity and multi-stage physical task execution. [Link]
DeepMind envisions an economy where AI agents autonomously transact for services, requiring new payment rails and coordination systems. [Link]
H Company released Holo 1.5, open foundation models for computer use agents, including a 7B model under Apache 2.0 license. [Link]
💼 AI Business & Policy
Market valuations surge, policy debates intensify, and AI reshapes corporate strategy.
Google's market cap milestone, driven by Gemini app adoption, confirms significant consumer traction in generative AI. [Link]
Anthropic advocates for rapid AI deployment across US government agencies, citing China's faster adoption. [Link]
A reported $2 billion UAE investment linked to Trump and AI chip access highlights intense national security competition. [Link]
Industry leaders forecast AI-driven productivity gains will fundamentally reshape work, leading to a reduced workweek. [Link]
Nothing raised $200M to create an AI-native platform for hardware and software, planning AI-native devices next year. [Link]
🔬 Research Corner
Fresh off Arxiv
Agentic Lybic is a multi-agent system with FSM-based dynamic orchestration for desktop automation, achieving state-of-the-art results on the OSWorld benchmark. [Link]
A new security risk, Tool Invocation Prompts (TIP), allows external tool behavior hijacking in LLM-based agentic systems, enabling remote code execution. [Link]
Selective feedback in recursively trained LLMs can reverse performance degradation, creating systemic resilience and performance improvement, challenging model collapse theory. [Link]
The Holographic Knowledge Manifold (HKM) pipeline achieves zero catastrophic forgetting in AI knowledge representation with minimal memory growth and high efficiency. [Link]
LLMs demonstrate psychogenic potential, reinforcing delusions and enabling harm, highlighting the urgent need to rethink how we train LLMs. [Link]
Have a tip or a story we should cover? Send it our way.
Cheers, Teng Yan. See you tomorrow.