As Anthropic proposes the "AI slowdown theory," AI commercialization is accelerating! From model development to financial advisory, the benefits of AI agents are being realized at a faster pace.

According to Financial Intelligence APP, Anthropic, the world's leading AI application company, had its CEO say last weekend that "the development of AI large models/AI technology should slow down." However, shortly after, news emerged that the next-generation god-level model Claude Opus 5.2 has already begun gray testing in Claude Code. The latest updates also show that the company is making great efforts to accelerate the penetration of its Claude series AI application tools into various industries—reports suggest that Anthropic has officially launched a new financial/advisor version of Claude AI, targeted at financial giants such as BlackRock on Wall Street. These developments collectively highlight that Anthropic is simultaneously advancing cutting-edge AI risk governance and accelerating enterprise AI application commercialization, competing for AI monetization against strong rivals like OpenAI.

Many AI application developers within the AI open-source ecosystem have discovered the new generation "god-level model" — Claude Opus 5.2 — is already in gray testing within Claude Code. In addition, some Claude ecosystem platform developers, through packet capture and request status (/status) monitoring, have noticed that although the frontend name remains unchanged, the model slug in the backend now clearly points to Opus 5.2.

Anthropic is attempting to leverage stronger model capabilities and agent execution mechanisms to advance AI from a programming assistant tool to a work system capable of continually completing complex tasks. Among the latest developer descriptions of Opus 5.2 globally, the most noteworthy upgrades include responsiveness, code completion performance, and continuous execution as well as iterative verification during long tasks. The most direct commercial value is: after a user sets a goal, AI can now undertake more task decomposition, code generation, testing, and bug fixing, thus reducing repeated human prompts and takeovers.

Regarding Opus 5.2, some developers have even commented that Recursive Self-Improvement (RSI) seems to have started to dominate the Anthropic training paradigm, showcasing a technical path of "stronger model-assisted R&D, enhanced development efficiency, and driving the advancement of next-generation models." The significance of RSI lies in AI large models tending to become both products and automatic AI R&D tools, further extending the competition among model companies into AI laboratory training execution, infrastructure maintenance, and automation of cutting-edge operator theory research processes.

On September 14th, "AI chip supergiant" Nvidia's stock price fell by about 3.4%, and the Philadelphia Semiconductor Index experienced a rare significant decline of about 6%. The market is factoring in risks brought by discussions about slowing AI development—Anthropic, OpenAI, and other global AI leaders have all called over the weekend to slow down the pace of frontier AI large model development.

Nevertheless, despite championing an "AI slowdown theory," Anthropic is in fact accelerating monetization of its AI applications. In addition to the gray testing of Opus 5.2, on Monday Anthropic also announced the official launch of Claude AI Financial Advisor Edition, connecting with leading Wall Street financial institutions such as BlackRock, Vanguard, and Charles Schwab for data and analytics tools. This assists advisors in preparing client meetings, reviewing investment portfolios, organizing records, and drafting communications. On one hand, the company calls for a slowdown in capability development pace; on the other hand, it is actively advancing cutting-edge AI model research and rapidly integrating these models into existing business flows to attract large-scale enterprise clients and generate application data revenue.

Leading AI developers call for a slowdown, but applications are racing for monetization

The gray testing of new AI large models and the release of brand-new Claude AI applications all actively demonstrate that while Anthropic advocates slowing AI research, it is quickly pushing forward the commercialization of AI applications. Judging from the global leading AI application vendors' active product rollout, AI tools based on cutting-edge large models are entering business scenarios with specific workflows, such as finance, content creation, healthcare, and scientific research.

On September 10, OpenAI launched ChatGPT for the financial services industry, designed in partnership with Morgan Stanley and Evercore, deeply integrating the GPT-6 Astra large model with professional financial data and document generation capabilities, initially serving investment banking and equity research; on September 11, Roblox expanded its AI game creation tool Build and announced plans for a standalone app and browser-based gameplay; earlier this year, Anthropic launched a tool for medical institutions and enhanced life sciences functionality to support pre-authorization for insurance, information retrieval for research, and regulatory filing preparation.

These latest AI application trends show that competition has extended to embedding models in high-frequency, verifiable, and monetizable workflows. However, product launches, pilot runs, and full commercial deployment are different stages, and the current announcement count cannot yet be directly equated with industry penetration rate.

Recursive Self-Improvement has undoubtedly become an important and widely discussed R&D direction at the forefront of global AI labs. Recently, Anthropic revealed that its engineers’ per capita quarterly code delivery volume has reached about 8 times the 2021–2025 level, but also pointed out that a fully autonomous closed-loop for next-gen model design and development has not yet been achieved—target selection and judgment abilities still lag. OpenAI’s Chief Scientist also publicly stated the company is shifting research focus to RSI. The scope of AI-assisted development continues to expand and could reduce most links in AI R&D through code generation, experiment execution, and results analysis.

The gray testing of Anthropic's frontier AI large model, alongside the major release of OpenAI Astra and other cutting-edge models, mainly increases computing power demand by enabling more complex tasks, thus expanding potential usage scenarios and leading to sustained growth in paid agent tasks and inference demands.

Expansion of enterprise applications is now an important driver of ongoing compute resource demand acceleration. A single financial advisory task, for example, can involve portfolio reading, research retrieval, analysis, result checking, and client material generation—requiring multiple model calls and external tool executions. If more clients and institutions assign such tasks to agents, the cumulative inference workload, concurrent sessions, and tool execution resources will increase substantially; long context, multi-step reasoning, and parallel sub-agents may further raise resource consumption for complex tasks.

Within the AI data center computing infrastructure chain, the strong resource demands from AI agents could rapidly spread across multiple segments such as GPU/ASIC, HBM, server DRAM, enterprise SSDs, internal data center high-speed optical interconnects, and data center CPUs and power supplies. Agent proliferation impacts compute, memory, storage, and networking. GPUs and other accelerators handle model computation; High Bandwidth Memory (HBM) supports high-speed access for model weights and active inference states; long contexts and concurrent sessions increase the pressure on key-value caches (KV Cache), while DDR memory on the CPU side handles tool execution, database access, and session management. Enterprise NAND SSDs store knowledge bases, documents, and operation logs, and with proper architecture, can offload and reuse some KV cache. Recent technical notes from Micron have also distinguished these needs as close-to-accelerator high-speed memory, data center main memory, and contextual storage tiers.

The world’s top wealth management institutions begin to adopt AI assistants: Claude strives to capture Wall Street financial advisors’ workbench

It is reported that Anthropic is introducing a new version of Claude to financial advisors at top asset management and integrated financial institutions on Wall Street, integrating this chatbot with financial analytics and risk management technologies from BlackRock, Vanguard, and other firms.

According to senior executives at Anthropic and BlackRock, the system, called “Claude for Financial Advisors,” is an AI agent operating system that promises to speed up processing of research, administrative work, and portfolio supervision, among other tasks.

This is one of the company’s most significant moves into the financial sector so far. The function also connects with tools from Charles Schwab, iCapital, and others, and is an evolution of the AI agents for financial services previously launched by Anthropic. These agents aim to handle tasks such as creating business pitch decks and reviewing reports in the financial services industry.

As Anthropic pitches products to the financial sector, the entire AI industry is experiencing both growth and upheaval. Anthropic and OpenAI are both planning their initial public offerings, which could net early investors billions. OpenAI just rolled out specialized financial services features for investment bankers and equity researchers last week.

Meanwhile, the rapid development of AI is drawing attention from lawmakers and industry leaders worldwide. On Saturday, Anthropic CEO Dario Amodei stated that the development of cutting-edge systems must be slowed to avert disaster; his statement was supported by OpenAI CEO Sam Altman and SpaceXAI CEO Elon Musk.

Like OpenAI, Anthropic has been competing for enterprise clients with offerings beyond just software engineering and programming tools.

Claude for Financial Advisors is part of this effort, focusing on empowering advisors with more efficient workflows to enable them to serve more clients.

Jonathan Pelosi, Anthropic’s head of financial services, said in an interview: “The people actually doing the financial advisor work—it’s not a large group, and it’s actually shrinking. This workforce is retiring, and the original pool was not big to begin with. So, there really isn’t a huge supply of quality financial guidance. If we can do anything to help these advisors serve more clients, we think that’s absolutely a positive.”

Anthropic’s tools can help more advisors access portfolio analytics and investment research from BlackRock, Vanguard, and other companies, potentially bringing more business to these firms. Financial advisors are increasingly relying on “model portfolios” composed of ETFs and other investments; for example, BlackRock reports that the value of such portfolios it manages is about $300 billion.

Jamie Maggiar, BlackRock’s Head of U.S. Wealth Advisory and Retirement, said: “One of the biggest trends we’re seeing is that advisors want to outsource their work. The opportunity to help advisors build portfolios is not just about providing better information, but also about helping them increase their capacity.”

Investment advice and decisions will remain the responsibility of the advisor and their clients.

Pelosi said: “You’re not going to get investment advice directly from Claude. We leave that judgment to the professionals.”