Raising 100 Lobsters ≠ Enterprise-level AI Agents?

Since the start of the year, the "lobster" AI trend has taken off, creating quite a buzz in the enterprise-level AI Agent sector.

Lobster AI has enabled many corporate decision-makers to envision, for the first time, the genuine potential for AI Agents to handle complex tasks from start to finish. Sun Zhenya, a research manager at IDC China, notes that this has significantly spurred the adoption of AI Agents in businesses, including the development of enterprise-level AI digital employees.

This trend has led to a significant surge in token usage among model providers, "consuming over one-sixth of global computing power," with rumors indicating that some collaboration platforms have exceeded their sales targets by 200%–300%. However, industry insiders informed Digital Frontier that this has not yet resulted in large-scale procurement orders in the enterprise productivity market.

“Personal-use ‘baby lobsters’ are not the same as the enterprise-level ‘lobsters’ that corporate clients require. Current ‘baby lobsters’ cannot truly deliver stable, secure, and accountable job-level work within enterprises,” stated Hua Luke, Chief Commercial Officer of Linghe Shuzhi, a vertical AI Agent provider for manufacturing scenarios. Enterprise-level digital employees must integrate into corporate processes, have defined roles and permissions, understand tasks and outputs, connect with various enterprise IT systems, and undergo final result verification. Additionally, enterprise clients highly value the service capabilities of AI companies, concerned about their ability to continuously assist enterprises in uncovering more valuable and transformative scenarios.

However, it's undeniable that the lobster AI trend has accelerated the development of enterprise-level AI Agent products. From large corporations to RPA vendors to AI Agent startups, there has been a steady stream of new product announcements recently. IDC previously forecasted that by 2030, the number of active AI Agents would rapidly increase from approximately 28.6 million in 2025 to 2.216 billion, with the number of active AI Agents more than doubling annually on average.

This trend is set to reshape the entire enterprise productivity market.

01

“No Large-Scale Enterprise Procurement Yet”

Hua Luke from Linghe Shuzhi received numerous inquiries from manufacturing clients around the Spring Festival, with most lobster-related inquiries coming from the second-generation successors of manufacturing companies. “Chinese manufacturing has reached a stage where most sizable enterprises are now 30 to 40 years old and are facing succession issues. Most of these successors have studied abroad, are curious about new technologies, and are eager to use advanced AI tools to assist in business operations and succession.”

Enterprise-level AI Agent providers like ShiZai AI, which serves operators, e-commerce, and energy industries, have also received numerous inquiries about lobster AI deployment from clients in recent months.

These are the ripples in the enterprise-level AI Agent market following the popularity of Openclaw.

OpenClaw is positioned as a 24/7 full-time digital employee, and raising a “lobster” is akin to having an employee who can manage computer work anytime, anywhere. This has put various enterprise-level AI Agent companies, which specialize in “enterprise-level digital employees” and emphasize the value of AI productivity, on high alert.

Before clients fully realized it, these companies took proactive steps internally. Ouyang, the head of algorithms at ShiZai AI, told Digital Frontier that they recognized the value of Openclaw when it was still called Clawdbot, and the technical team immediately convened a dedicated analysis meeting.

“Its end-to-end automation execution stands out, filling the gap left by Manus in its inability to manipulate and call local system functions in the cloud, and its integration with IM tools is a significant advantage,” Ouyang mentioned. ShiZai AI drew inspiration from Openclaw’s design, and the product development team expedited the internal testing and launch of the “ShiZai Agent Boundless Edition” based on their existing auto-execution robots. This new version supports embedding into IM software like DingTalk, enabling cross-device wake-up and control similar to Openclaw.

Linghe Shuzhi, which focuses on vertical digital employee platforms for manufacturing, also launched its own community, Linkclaw, akin to Moltbook overseas, shortly after Moltbook's debut. Hua Luke from Linghe Shuzhi told Digital Frontier that they noticed the rapid growth of lobster AI as soon as it emerged.

Since Linghe Shuzhi has been developing AI digital employee platforms for manufacturing, the interactions between AI Agents in overseas lobster communities have given them new ideas for creating a “Linghe Factory” for manufacturing. “Similar to Stanford’s small town, placing AI Agents in different roles within a factory where they can collaborate and understand each other could potentially lead to more innovations.”

Some AI system integrators deeply rooted in the industry are also leveraging Openclaw's capabilities to upgrade their product offerings. For example, Zhongguancun Kejin launched PowerClaw, an enterprise-level solution for OpenClaw, and Fengqing Technology (Fabarta) introduced lobster-like products.

After enterprises took proactive steps themselves, the consumer market also heated up simultaneously. Reports indicate that the global daily new deployment instances surged from 5,000 to 90,000, an 18-fold increase. The enthusiasm in the consumer market has also spread to the client base of enterprise-level AI Agents, with multiple AI Agent vendors reporting receiving client inquiries seeking advice on lobster AI deployment.

However, “these have not translated into enterprise-level procurement.” Ouyang admitted that most inquiries remain at the individual discussion level, with no enterprises currently having the budget or explicitly proposing to purchase similar systems.

Hua Luke also believes that the lobster AI trend is more about educating the market and raising expectations. “Penetrating manufacturing enterprises with a new product usually takes a long time and significant educational effort. With the lobster AI trend, many business owners have learned about terms like AI Agent, Skill, and Workflow for the first time. They are high-quality seed users for enterprise-level lobster AI. Although they are anxious and mostly observing from the sidelines, since the penetration rate of AI in Chinese manufacturing is less than 5%, they will soon focus on how AI Agents can truly create value within enterprises.”

Entrepreneurs who have started businesses based on lobster AI at various lobster-themed events also admit that currently, even if some enterprises are quick to adopt lobster AI, they tend to focus on one or two innovative business scenarios, with few bosses daring to open up core business processes to lobster AI. Multiple enterprise-level digital employee platforms have mentioned that lobster AI is more suitable for one-person companies without complex business logic.

02

Implementation Challenges: Security, Cognitive Divergence, Lack of Skills, ROI

Lobster AI has quickly unified consumer awareness, forming a deployment trend. Why hasn't this progressed quickly in the enterprise market?

Outsiders believe this is related to factors such as security and compliance requirements, enterprise procurement decision-making chains, lack of unified awareness, and ROI measurement.

Security and controllability are, of course, core concerns. Ouyang’s ShiZai AI serves many state-owned enterprises, whose clients are highly concerned about data leakage, model controllability, and operational security. The open-source nature of lobster AI and its reliance on public large models make many enterprises hesitant.

At the same time, lobster AI is considered to have some distance from true enterprise-level digital employees. IDC's Sun Zhenya told Digital Frontier that the key lies in whether they can manage a complete closed-loop process for a role, becoming a digital workforce capable of completing tasks with a feasible ROI. This requires business scenario knowledge, deep integration with the enterprise's own systems, data, and processes, as well as permission boundaries, behavioral auditing, and cost governance.

Additionally, from the procurement decision-making chain perspective, although lobster AI has been popular on the consumer side, there is currently no consensus in the industry that “enterprise-level AI Agents = lobster AI.” “Last year, a large number of clients purchased orchestration-type AI Agents, which are still being implemented and digested,” Ouyang said. This has also influenced procurement decisions at the enterprise level.

Especially for state-owned and central enterprise client groups, their understanding of AI Agents is influenced by policy guidance. Enterprises focus on coverage metrics for AI Agents. At the same time, this client group does not necessarily believe that all AI Agents need autonomous planning capabilities based on large models like Openclaw, using tools to complete tasks. Configuration-type AI Agents formed through orchestration tools and workflows may offer stronger certainty for them.

“For these enterprises, orchestration-type AI Agents provide a very clear process. If an error occurs in one step, they can identify which node went wrong and correct it. However, with lobster AI, the results of each call may differ,” Ouyang said. This may make lobster AI and similar AI Agents not the first choice for government and enterprise clients in the short term.

Earlier this year, IDC made a similar judgment about the enterprise-level AI Agent market in a report. IDC divides global enterprise-side AI Agents into four categories, with the currently popular lobster AI classified as independent AI Agents. As ecological standards and business models mature, independent AI Agents will gradually grow into one of the mainstream forms for handling complex tasks and cross-system collaboration. However, for now, the report suggests that configuration-type AI Agents are currently the mainstream way for enterprises to quickly build AI Agent capabilities. In the medium to long term, configuration-type AI Agents are expected to become the most widely deployed type on the enterprise side, providing long-tail capability supplementation for enterprises.

Currently, independent AI Agents still have a long way to go. “The hottest thing in the Skills market is finding Skills for Skills, followed by various web search Skills,” a classic joke reflects the current lack of Skills for genuine enterprise application scenarios under the open-source ecosystem, with various enterprise-level scenarios, capabilities, and ecosystems yet to be further connected.

Sun Zhenya believes that the current lobster AI trend proves that the capability ceiling for AI Agents is higher than most people expected but also exposes that the market's preparedness is lower than anticipated. “The gap in between represents the biggest business opportunity for the future enterprise-level AI Agent market.”

Meanwhile, some industry insiders believe that another pressing question for the current implementation of enterprise-level AI Agents is cost and ROI measurement.

During the personal lobster AI deployment trend, many mentioned the substantial token consumption, with a single task potentially exhausting a month's token quota. It is foreseeable that when AI Agents are deployed in production scenarios, enterprises will inevitably focus on operational costs to ensure sustainable ROI while achieving business value.

However, compared to the consumer market's concern over token consumption metrics, the enterprise market is more focused on the business value generated by AI Agents. “Converting models and tokens into genuine job-level deliverables required by enterprises can be decoupled from token consumption and instead linked to output,” Hua Luke from Linghe Shuzhi mentioned. If based solely on token consumption, low-frequency, high-value scenarios might be underestimated in commercialization. However, if linked to job-level output, low-frequency, high-value scenarios can also form a strong commercialization path.

Regarding ROI measurement and scalable value, the industry generally believes that enterprise-level AI is still in the early stages of value realization. Currently, implementable enterprise-level AI tends to focus on high-value, easily measurable ROI scenarios, such as customer service, software development, financial processes, and supply chain optimization.

Taking manufacturing as an example, Hua Luke from Linghe Shuzhi mentioned that they choose to prioritize AI Agent implementation in high-cost scenarios closely related to cost reduction in manufacturing, such as quality management, scheduling, and supply chain management. “Besides market education, the arrival of lobster AI has also prompted us technically to leverage existing technologies in the market rather than building everything from scratch,” Hua Luke believes this will accelerate their R&D process. This year, they hope to expand the number of manufacturing AI Agent scenarios from over a dozen to 100.

03

Major Restructuring of Enterprise Productivity: How to Integrate Multi-AI Agent Systems

In the short term, the implementation of independent AI Agents like lobster AI will still take time. However, the capabilities demonstrated by lobster AI—cross-system calls, autonomous planning, and tool integration—have educated the market and shown the possibility of a fundamental transformation in work patterns within enterprises.

A significant difference here is the emergence of multi-AI Agent systems within enterprises. Hua Luke from Linghe Shuzhi uses a beehive as a metaphor for enterprise needs. A single bee is not particularly intelligent, but a beehive has a collaborative system. Currently, “raising 100 lobsters does not equate to having swarm intelligence.” Enterprises need to manage AI Agents, enabling them to collaborate vertically, control permissions and goals, and ensure long-term usability.

This necessitates a control and coordination platform above the various AI Agents within enterprises. New roles are emerging.

The leading venture capital firm in Silicon Valley, a16z, has recently forecasted that the Enterprise Multi-Agent Orchestration Layer will emerge as a key trend in the enterprise AI market by 2026. Businesses will require a novel “coordination system” to oversee multi-Agent interactions, mediate context, and guarantee the dependability of autonomous workflows. This shift will give rise to new roles within companies, ushering in an entirely novel mode of workflow operation within large-scale organizations.

This also implies that, regarding the utilization of Agents within enterprises, various suppliers must redefine their roles—whether to offer productivity tools tailored to specific scenarios or professions or to position themselves as a scheduling and coordination layer within enterprises. This will represent a significant reorganization of enterprise productivity.

Collaboration platforms such as Feishu and DingTalk are evolving into enterprise-grade AI-native work platforms. Given their integration into daily enterprise work environments and possession of essential infrastructure like organizational structures, permission systems, and data interfaces, they are ideally suited to serve as Agent coordination hubs for millions of enterprises. However, this positioning comes with both significant advantages and limitations—they boast substantial traffic advantages but may lack in-depth industry-specific capabilities.

Among the competitors, DingTalk has made more assertive strides. A week ago, Alibaba introduced Wukong as a unified portal for Alibaba's AI capabilities in enterprise work settings, which will be seamlessly integrated into DingTalk. It will support connections to users' DingTalk accounts, security access permissions, and enterprise application systems. Additionally, it features an AI capability marketplace, facilitating future third-party skill transactions, which will enable it to serve a broader range of industry enterprises.

Vertical industry Agent providers have ventured into complex business productivity scenarios characterized by intricate business logic. This includes suppliers offering comprehensive vertical digital productivity solutions and those specializing in a single vertical productivity domain.

Suppliers providing comprehensive vertical digital employee solutions emphasize multi-Agent collaboration platforms and enterprise management capabilities for Agents. Taking Linghe Shuzhi as an example, its digital employee platform includes a talent marketplace where enterprises can engage AI employees for vertical scenarios as needed, a trusted space for accumulating business experience and organizational knowledge, and an entry point for human-machine collaboration within enterprises. Large enterprises can develop their private skills based on this platform, amassing unique knowledge and experience within the company. For these solution providers, Longxia (a general-purpose AI platform) complements their vertical job-level Agents—general personal-oriented outputs are managed by Longxia, while company core objectives are entrusted to job-level Agents.

Some suppliers previously focused on vertical job-type Agents are exploring integration into the open ecosystem of Longxia.

These explorations are geared towards proactively adapting to the market's transition from human-oriented to agent-oriented products. Naturally, companies are also monitoring the subsequent impacts. Analysis indicates that the new ecosystem has the potential to disrupt original business models based on product interfaces, posing a substantial challenge to vertical vendors in terms of their scenario understanding and data barriers.

In this wave, tool vendors with more general capabilities may face the greatest pressure. Just as Jasper was affected by OpenAI's own offerings, in marketing scenarios, some practitioners have already constructed workflows for public opinion monitoring and brand insights directly on Longxia, bypassing the original marketing AI employee products.

Moreover, collaboration vendors, with their vast user bases, are also introducing scenario-based solutions for general job roles. For instance, Alibaba Wukong has unveiled solutions for ten major OPT scenarios, including e-commerce, cross-border e-commerce, knowledge-based blogging, development, stores, design, manufacturing, legal, finance and taxation, and headhunting. Players lacking vertical depth may need to reconsider their positions and competitive advantages.

Longer-term questions also remain unanswered. With the further evolution of various job-level agents and the gradual establishment of multi-agent systems, is there still a need for products designed for human use? In the future, will vertical vendors become part of the ecosystem of collaboration vendors, or will the middle layer vanish as agent intelligence advances, with agents directly connecting to backend APIs and databases?

Amidst this intricate landscape of competition and cooperation, numerous uncertainties persist. This will be a tumultuous restructuring, and further changes are yet to unfold.