Artificial intelligence applications are progressing into a critical next phase, moving beyond developer tools to achieve widespread adoption among knowledge workers. A recent Goldman Sachs research report outlines a clear path for the scalable implementation of general-purpose AI tools, using two landmark products—Claude Cowork and OpenClaw—as key examples.
According to the report, the next stage of AI application as a general intelligence layer is becoming evident, with Anthropic's Claude Cowork and the open-source agent OpenClaw serving as two core data points defining this trend.
The report highlights Claude Cowork as the first credible case of an "agent workflow" tailored for knowledge workers, which is reshaping value distribution within the SaaS industry. Rather than directly replacing existing systems of record such as CRM or ERP software, it is shifting market share toward SaaS providers that can effectively integrate AI capabilities.
Simultaneously, the viral spread of OpenClaw not only demonstrates a new form of future human-computer interaction but its unique hosting economics are challenging the dominance of hyperscale cloud providers. This trend is redirecting massive new inference computing demands toward alternative edge computing and Virtual Private Server (VPS) providers like DigitalOcean and Cloudflare.
Claude Cowork: The First Credible Agent Workflow for Knowledge Workers On January 12, Anthropic officially launched Claude Cowork, an agent tool embedded within the Claude desktop application. Designed specifically for non-developers, it automates file management and task processes involving multi-step operations, such as file organization, document processing, and workflow management.
By January 30, Anthropic had further released a series of Cowork plugins targeting various "business functions," covering scenarios including sales and marketing, finance, legal, customer support, data analysis, productivity, enterprise search, and product management.
Goldman Sachs believes Cowork's core value lies in bridging a critical gap between developer-specific agent tools and the chat-based interfaces typically used by general knowledge workers. While software engineers have benefited from productivity gains via terminal agent loops for over a year, AI interaction for knowledge workers has long remained confined to chat interfaces. Cowork extends this capability to non-technical personnel through an application interface that integrates with tools like Excel and standard business software.
Specific implementation scenarios include accountants categorizing credit card transactions, HR teams compiling tax reports and tracking paid time off (PTO), and CFOs accelerating project-specific Discounted Cash Flow (DCF) analysis. Among digital agent customers, marketing applications are particularly prominent, with clients integrating platforms like Klaviyo to automate entire reporting processes.
Cowork's core competitive advantage is its function as an intelligent layer across the SaaS stack. It is not dependent on any single SaaS product, can switch freely between tools, and performs reasoning on data spanning multiple systems of record and "glue layer" applications. Industry research by Goldman Sachs indicates that Anthropic holds a significant lead in both tool capabilities and the underlying model's intelligence.
Cowork Won't Replace Systems of Record, But Will Reshape SaaS Value Distribution Goldman Sachs clearly states that agent tools like Cowork will not replace core systems of record like CRM or ERP, and enterprises show no inclination to dismantle their existing SaaS stacks. However, as AI becomes a key differentiator, spending will increasingly concentrate on SaaS vendors that can deliver the most effective AI workflows.
In practice, customers are using Cowork to replace lightweight "glue" software—tools primarily used for transferring data between systems—rather than core platforms. Discussions with Goldman Sachs' partners indicate that companies still prefer paying third-party software vendors for security, governance, and operational control.
The partnership between Intuit and Anthropic serves as a prime example of this logic. On February 24, Intuit announced a collaboration with Anthropic: Intuit will gain access to Anthropic's models and custom AI agents via the Claude Agent SDK, while Anthropic will promote Intuit products on its Cowork, Claude for Enterprise, and Claude.ai platforms through MCP integration.
Goldman Sachs views the rationale behind this partnership as similar to Intuit's previously reported $100 million annual agreement with OpenAI: the foundational model company monetizes its model capabilities and reach without building custom workflows, while Intuit retains control over its data and maintains its unit economics.
A potential risk identified is that inserting an additional layer between the customer and Intuit could lead to value abstraction over the long term. However, a potential benefit is improved customer acquisition, akin to partnership models with search engines during the Web 2.0 era.
Notably, on March 5, OpenAI launched GPT-5.4, its first general-purpose model with native computer usage capabilities. Similar to Claude Cowork, it can operate a computer and execute agent workflows across applications. OpenAI specifically highlighted that this model reduces token consumption by approximately 47% on certain tasks compared to its predecessor, while maintaining the same level of accuracy.
OpenClaw: Breaking Context Limits and Demonstrating New Directions for Interaction and Compute Demand Goldman Sachs posits that OpenClaw has achieved viral growth in consumer applications by granting AI agents full computer control and unlimited memory. Its unique local and low-cost VPS hosting model is proving to be a significant boon for edge computing providers.
OpenClaw (formerly known as Moltbot and Clawdbot) was first released in November 2025 and experienced "viral" spread in late January. It is an open-source, self-hosted, always-on AI agent that runs on a user's local computer. It can receive instructions via personal messaging platforms like iMessage, WhatsApp, and Google Chat, and execute virtually any task a human can perform, including opening applications, browsing the web, and completing transactions.
OpenClaw addresses a key technical bottleneck limiting most AI agents: the context window problem. Its solution is a file-based memory system where all agent activities are stored in Markdown files on the local computer. The agent loads the day's memory and the previous day's context at the start of each session. Users can also establish long-term memory through instructions (e.g., "always check OpenTable first when booking a restaurant"), creating a persistent preference database.
Matthew Prince, Co-founder and CEO of Cloudflare, suggested that OpenClaw's development trajectory over the next three years might mirror ChatGPT's path over the past three years. It is currently catalyzing significant experimentation, but similar to ChatGPT's early days, it might take several years to achieve sustained, scalable daily usage.
OpenClaw's Compute Economics: A Structural Opportunity for Edge and VPS Providers OpenClaw's hosting economics offer important insights for investors. Due to cost volatility associated with hyperscale cloud providers, most users opt to run OpenClaw on home hardware (like Mac Minis or old laptops) or low-cost VPS providers such as Cloudflare and DigitalOcean. Some users reportedly incurred token costs reaching thousands of dollars within the first few days of use.
Many users are adopting a layered model architecture where a frontier model handles orchestration, and smaller models execute sub-tasks, thereby reducing the cost per query.
Data from DigitalOcean provides a clear illustration: within days of OpenClaw's release, customers deployed nearly 30,000 native one-click OpenClaw Droplets on DigitalOcean, with thousands of additional deployments activated. Within DigitalOcean's AI customer revenue, the proportion from inference services continues to grow. As of December 2025, AI customer Annual Recurring Revenue (ARR) reached $120 million, with non-bare metal services accounting for 70% of this, and inference services growing 254% year-over-year.
While Cloudflare has not yet observed a significant impact from OpenClaw on customer growth or network traffic, Goldman Sachs believes this trend further validates the leading price-performance advantages of Cloudflare's edge network and Workers software.
Comments