Goldman Sachs’ judgment: Claude Cowork becomes the first “trusted case for agent workflow,” OpenClaw demonstrates the future direction of human-computer interaction

AI applications are entering a critical next phase—comprehensive penetration from developer tools to knowledge workers. Goldman Sachs' latest research report outlines a clear path for the scaled deployment of general intelligence tools through two iconic products—Claude Cowork and OpenClaw.

On March 7, according to ZF Trading Desk, Goldman Sachs pointed out in its latest research report that the next phase of AI as a general intelligence layer has already emerged, and Anthropic’s Claude Cowork and the open-source agent OpenClaw are defining this trend as two core data points.

The report states that Claude Cowork, as the first "trusted agent workflow case" for knowledge workers, is reshaping the value distribution in the SaaS industry—it will not directly replace existing systems of record (such as CRM or ERP), but will shift market share towards SaaS providers who can effectively integrate AI capabilities.

Meanwhile, the viral spread of OpenClaw not only showcases a new form of human-computer interaction, its unique hosting economics are breaking the monopoly of hyperscalers, directing the massive new demand for inference computing power to alternative edge computing and VPS (virtual private server) providers such as DigitalOcean and Cloudflare.

Claude Cowork: The first trusted agent workflow solution for knowledge workers

On January 12, Anthropic officially launched Claude Cowork—a proxy tool embedded in the Claude desktop app, designed especially for non-developers, able to automate file management and task processes, covering multi-step operations such as file sorting, document processing, and workflow management.

On January 30, Anthropic further released a series of Cowork plugins for "business functions," covering scenarios in sales and marketing, finance, legal, customer support, data analysis, productivity & enterprise search, as well as product management.

Goldman Sachs believes Cowork’s core value lies in filling the critical gap between developer-specific agent tools and everyday chatbots used by ordinary knowledge workers. Software engineers have enjoyed productivity gains through terminal agent loops for over a year, while AI interaction for knowledge workers has long remained at the chat interface level. Cowork extends this capability to non-technical personnel via an application interface integrated with Excel and standard business tools.

Specific application scenarios include: accountants categorizing credit card transactions, HR teams compiling tax reports and tracking PTO, CFOs accelerating project-specific DCF analysis. Among digital agent customers, marketing scenarios are most prominent—customers are integrating Klaviyo and automating the entire reporting process.

Cowork’s core competitive advantage is its role as an intelligent layer across the SaaS stack: it is not attached to any single SaaS product, can freely switch between tools, and can reason about data that traverses multiple systems of record and "glue layers." Industry research by Goldman Sachs shows Anthropic is clearly ahead in both tool capabilities and underlying model intelligence.

Cowork will not replace systems of record, but will reshape SaaS value distribution

Goldman Sachs makes it clear that agent tools like Cowork will not replace core systems of record such as CRM or ERP, and enterprises do not intend to dismantle existing SaaS stacks. However, as AI becomes a key differentiating factor, wallet share will concentrate on SaaS vendors with the most effective AI workflows.

In practice, customers are using Cowork to replace lightweight "glue" software—tools whose main function is moving data between systems—rather than core platforms. Conversations with Goldman Sachs partners show enterprises still prefer to pay third-party software vendors for security, governance, and operational control.

The Intuit–Anthropic partnership is the best illustration of this logic. On February 24, Intuit announced a collaboration with Anthropic: Intuit will access Anthropic models and custom AI agents via the Claude Agent SDK, and Anthropic will use MCP integration to promote Intuit products on Cowork, Claude for Enterprise, and Claude.ai platforms.

Goldman Sachs believes this logic is similar to the $100 million-a-year agreement Intuit previously signed with OpenAI: the base model does not require custom workflows, but monetizes through model capabilities and reach; Intuit maintains control of its data and preserves unit economics.

Goldman Sachs points out that the potential risk is that adding a layer between customers and Intuit may abstract value long-term; but the potential reward is better customer acquisition, akin to the partnership model with search engines during the Web 2.0 era.

Notably, on March 5, OpenAI launched GPT-5.4—its first general model with native computer operation capability. Like Claude Cowork, it can operate computers and execute agent workflows across applications. OpenAI especially highlighted that the model reduces token consumption by about 47% versus previous models for some tasks, while maintaining identical accuracy.

OpenClaw: Breaking context limitations, showing new directions for human-computer interaction and computing power demand

Goldman Sachs believes OpenClaw achieves viral growth in consumer applications by giving AI agents full computer control and infinite memory; its unique local and low-cost VPS hosting model is proving a major boon for edge computing providers.

OpenClaw (formerly Moltbot and Clawdbot) was first released in November 2025, and achieved "viral" spread by late January.

This is an open-source, self-hosted, always-on AI agent that runs on users’ local computers, can accept commands via personal messaging platforms such as iMessage, WhatsApp, and Google Chat, and perform any task a human can complete, including opening applications, browsing the web, and completing transactions.

OpenClaw solved the key technical bottleneck for most AI agents—the context window problem. Its solution is a file-based memory system: all agent activity is stored in local Markdown files, and the agent loads today's memory and the prior day's context at the start of each session. Users can also create long-term memory (such as "always check OpenTable first when booking restaurants") via commands, forming a persistent preferences database.

Cloudflare co-founder and CEO Matthew Prince pointed out that OpenClaw's next 3 years could replicate ChatGPT’s path over the past 3 years: it's currently catalyzing lots of experimentation, but like ChatGPT’s early days, may take years to achieve sustained, scaled daily use.

OpenClaw's computing economics: Structural opportunities for edge computing and VPS providers

OpenClaw’s hosting economics bring important insights for investors. Due to highly variable costs with hyperscalers, most users choose to run OpenClaw on home hardware (Mac Minis, old laptops) or low-cost VPS providers (such as Cloudflare and DigitalOcean). Some users' token usage reaches thousands of dollars within the first few days.

Many users are adopting a layered model architecture: frontier models handle orchestration, smaller models execute sub-tasks, thus reducing the cost per query.

DigitalOcean’s data is most direct: Within days of OpenClaw’s release, customers deployed nearly 30,000 native one-click OpenClaw Droplets on DigitalOcean, with thousands of additional deployments activated. For AI customers, inference services' share of revenue keeps growing—annual recurring revenue for AI customers in December 2025 hit $120 million, with non-bare-metal business making up 70%, and inference services’ revenue up 254% year-over-year.

Cloudflare has not yet seen a significant impact on logo growth or network traffic from OpenClaw, but Goldman Sachs believes this trend further validates Cloudflare’s edge network and Workers software's leading price-performance advantage.

~~~~~~~~~~~~~~~~~~~~~~~~

The above exciting content comes from ZF Trading Desk.

For more detailed interpretation, including real-time commentary and frontline research, please join【ZF Trading Desk▪Annual Membership】

Risk Disclosure and DisclaimerThe market involves risks, investments require caution. This article does not constitute personal investment advice, nor does it take into account special investment goals, financial situation, or needs of individual users. Users should consider if any opinions, perspectives, or conclusions in this article fit their specific situation. Invest accordingly, liability at your own risk.