Why has "raising lobsters" become a phenomenon? Which industries could "AI lobsters" disrupt, and how can they be commercialized? According to a China equity research report published by JPMorgan Chase on March 12, the bank's analysts recently held discussions with the project manager of KNOWLEDGE ATLAS's AutoClaw, providing a deep analysis of the reasons behind the popularity of products like AutoClaw and OpenClaw, as well as the subsequent paths and logic for application deployment and commercial monetization. JPMorgan Chase believes that "products like AutoClaw and OpenClaw are important not because they have suddenly made autonomous AI commercially viable, but because they have significantly lowered the barrier for non-technical users to experience agent workflows." The most critical impact for the market is this: while the adoption of agents is expected to increase model usage and infrastructure demand, short-term monetization remains in its early stages. Actual deployment will first occur in relatively structured workflows, rather than through broad, fully autonomous human replacement.
The "Lobster" Agent Surge: A Victory in Product Design, Not a Model Leap Is the recent fervor around OpenClaw-type products due to a leap in model capability or an optimization in product design? The interviewee's comments provided a clear answer. This popularity "reflects improvements in product design and usability, rather than a sudden mutation in model intelligence." The interviewee specifically highlighted three key factors: "integration with existing communication tools, persistent memory allowing agents to build user profiles over time, and broader system permissions that widen the practical scope of an agent's work." JPMorgan Chase notes that this distinction is crucial. The current trend is driven by better productization and accessibility, suggesting that user engagement can be expanded before achieving true enterprise-level monetization.
Foundation Model Quality: The Core Determinant of Commercial Ceiling In an era brimming with new agents, will foundation models become commoditized? The clearest point from the discussion was that "the commercial ceiling for agents still largely depends on the quality of the underlying foundation model." The interviewee repeatedly emphasized that "an agent is essentially a container or medium; the model remains the core factor determining whether tasks can be completed accurately and consistently, and whether the depth of reasoning is sufficient to function in high-value contexts." JPMorgan Chase views this as countering the notion that the agent layer will fully commoditize the model layer in the short term. "Better models should still translate into superior task completion, stronger instruction following, more stable long-context performance, and better handling of open-ended workflows." Therefore, increased agent adoption remains a significant positive for leading model suppliers.
Commercial Monetization: Still Early Stage, Focused on Structured Tasks Despite the hot concept of agents, the interviewee's remarks implied that "the agent market might still lean towards an exploratory phase without full monetization in the short term." Current products are still in the stage of helping users discover use cases. Significant expansion into commercial scenarios "might require another 6 to 12 months of model improvements, workflow training data, and product iteration." JPMorgan Chase believes this aligns with the current state of enterprise AI. "Coding and technical workflows remain the clearest early monetization path because the tasks are structured, the target functions are clearer, and the execution轨迹 are easier to define." Beyond coding, the lack of standardized "trajectory" data is a major factor limiting agents' ability to perform real-world, multi-step tasks.
Market Deployment: Technical Engineering and Structured Workflows Lead Which areas will adopt agents first as they go to market? The interviewee emphasized three broad categories: 1. Technical Engineering: "Expanding from coding to testing, deployment, operations, configuration, and debugging. This appears to be the most commercially credible category today." 2. Information and Content Workflows: "Including research, report generation, document processing, office file manipulation, and internal content creation." 3. Personal Productivity: "Such as email, calendar, and message management." While attractive to consumers, converting this into sustained monetization might take longer. JPMorgan Chase advises investors to "anchor short-term expectations on technical and structured enterprise tasks, rather than making overly aggressive inferences from consumer experimentation."
Open Architecture and Moat: Rapidly Replicable Features Are Not Key Another important view from the discussion was that the agent layer might not be a winner-take-all, proprietary channel for models. AutoClaw supports multiple model providers, with management explicitly endorsing an open architecture rather than mandating the use of only KNOWLEDGE ATLAS models. JPMorgan Chase believes this broadens the product's potential market and enhances the opportunity for the agent platform to become an aggregator within the model ecosystem. However, for model providers, this implies that "the agent interface itself may not guarantee exclusive downstream value capture, unless the supplier also leads in model performance, agent tool invocation, and workflow integration." Regarding competitive moats, management suggested that feature comparisons are less critical, as many visible features can be quickly replicated. They believe true defensibility is built on three aspects: "speed of product insight, foundation model quality, and cumulative agent capabilities (such as browser tools, memory systems, and workflow handling)." JPMorgan Chase concurs with this view, noting that investors should focus on whether suppliers can "consistently improve task completion rates, reduce friction, and leverage usage data to enhance agent performance over time."
Industry Chain Reshaping: Who Benefits, Who Gets Disrupted? Broader adoption of agents is expected to benefit multiple parts of the AI stack: 1. Firstly, it should benefit model suppliers, as more autonomous workflows imply greater token consumption and more persistent usage. 2. Secondly, it should benefit inference infrastructure, cloud, and related computing providers, especially if demand growth continues to outpace supply. 3. Thirdly, collaboration and workflow platforms that offer public APIs and allow controlled integration could also benefit by becoming the interface through which agents are embedded into daily work. Conversely, businesses whose value proposition relies on "shallow intermediation or low-barrier information processing" may face risks. AI is more likely to exert pressure on "roles or services with limited moats, publicly available information, and relatively easy-to-automate workflows." Furthermore, security and regulation are practical constraints for enterprise deployment. Management identified "prompt injection, permission errors, malicious third-party skills, and software vulnerabilities as practical constraints." This might slow monetization in the short term but increases the importance of trusted suppliers and compliance-grade architectures. JPMorgan Chase maintains an "Overweight" rating on KNOWLEDGE ATLAS, with a target price of HKD 800 until December 2026, based on 30 times the estimated 2030 price-to-earnings ratio, anticipating a compound annual revenue growth rate exceeding 100% from 2026 to 2030.
Comments