Full Summary
This Monday morning, OpenAI engineer Weng Jiayi has proposed a groundbreaking method for Agentic AI: self-modification of its own code, a development confirmed by both 36Kr and eu.36kr.com. Instead of relying solely on more data and computing power, Jiayi's experiment showed an AI, Codex, repeatedly writing, running, and modifying its strategy code to fix failures. This process led Codex to achieve a perfect score in Atari Breakout and perform well in robot control simulations, suggesting AI could evolve by continuously improving its engineering system. Meanwhile, Agentic AI is making significant strides in various sectors. Broadridge, for instance, has launched production-grade agentic AI in capital markets and wealth operations, with FinTech Global, Stock Titan, and Asset Servicing Times all reporting that new clients can see up to a 30% operational cost reduction from day one. These AI agents are already processing millions of transactions monthly across more than 40 clients, handling tasks from trade fails management to customer inquiry automation under human supervision. In retail, online businesses face a major shift as AI shopping agents begin making purchasing decisions, meaning retailers must now design for non-human customers, according to retailbiz. Stripe's Head of Solutions Architecture notes that more AI agents are interacting with their documentation than humans, highlighting their quiet emergence as first users of digital services. The AI Journal and it-online.co.za both explain that Agentic AI moves beyond simply generating information, allowing autonomous agents to execute complex tasks and deliver outcomes. This ranges from planning business trips to enhancing customer experience by understanding context and determining next steps. However, Digital Journal points out a significant "orchestration gap," where standalone AI agents struggle with complex, multi-step workflows, causing over 40% of agentic AI projects to fail. Solutions like 20X from Peakflo are emerging to address this by orchestrating tasks among specialized AI agents. This proliferation of agentic AI means your next online purchase might be made by a machine, and the financial services you use could be managed with significantly reduced operational costs, impacting efficiency and potentially the price of services.