The First Lobster Batch Deployment Management Tool: How Lobster, Which Raised Millions in Funding, Wants to Help Everyone Build a "Company of One"

"Some people are making tens of thousands a month off this." By Ren Qian

At 3 a.m., self-media blogger Ayu's 32nd Xiaohongshu account auto-published a post: "OMG! If I'd known this trick sooner, I would've saved at least 200 hours of overtime." She was fast asleep. Meanwhile, in the cloud, 36 "lobsters" were executing a battle plan: four back-end lobsters scraped trending topics, reverse-engineered viral hits, and screened for banned terms, while 32 front-end lobsters each embodied a different persona — "internet ops girl, three years on the job," "office worker who paid off 80K in mortgage through side hustles," "veteran PM with eight years' experience" — posting simultaneously across 32 vertical accounts.

Three months ago, Ayu was just another content creator making ¥8,000 a month, burned out from manually running five accounts. Now she commands a "content army" that produces 160 posts daily, and her income has quadrupled.

She's not alone. In Shenzhen, cross-border e-commerce seller Xiao Chen deploys 12 lobsters across four squads — reconnaissance, outreach, follow-up, and quoting — selling silicone products to North America and Southeast Asia. In Shanghai, Lao Zhang hires four lobsters to monitor investments, manage to-dos, and maintain his social calendar, stealing back ten hours a week to "let his brain clock out."

The tool they use is ClawTroop — a product recently launched by AI agent infrastructure company Moli Context, founded in 2025, now serving self-media, cross-border e-commerce, and enterprise productivity scenarios. ClawTroop is a deployment and management tool for mass "crayfish" (small AI agents), designed to lower the barrier to managing agent clusters so individuals and businesses alike can own, train, and orchestrate their own agent swarms at minimal cost.

"Dark Current Waves" has learned exclusively that Moli Context has secured a million-dollar investment from Hong Kong-based Data Intelligence Group Limited (DIG). Its founding team hails from Tencent, ByteDance, and other major tech firms, with graduates from Imperial College London, University of Glasgow, and University of Hong Kong among their ranks.

Anyone who's raised OpenClaw crayfish knows that deploying just one takes at least one to two hours of tinkering — let alone feeding, using, scheduling, and managing them in bulk. At the inflection point of the agent explosion, Moli Context is betting on an experiment in "management-first": if every worker will soon command an AI squad, the ability to write the rules of engagement may prove more valuable than individual combat skills.

Image source: Company provided

Part 01

Text as Military Doctrine

Before settling on the "Lobster Troop" direction, Moli Context's team stumbled on two earlier projects.

The first was Cowork, an MCP protocol-based tool aiming to build the ultimate personal productivity assistant. "Back then the MCP protocol was highly unstable, and the underlying model capabilities weren't smart enough," recalls Zhou Man, product lead. The second project was more ambitious — exploring whether VibeCoding, then more mature, could spawn a new interactive H5 medium beyond text, image, and video. They launched in North American and Southeast Asian markets, but customer acquisition costs were crushing and the two-sided market never reached viability.

The real turning point lurked behind the pain points. As heavy early adopters of AI tools, they began intensive use of OpenClaw to boost efficiency after its release. A paradox quickly emerged: a single agent delivered clear personal productivity gains, but getting an entire team onboard was "absurdly difficult." The long queues outside Tencent's offices to install crayfish offered further proof — the bottleneck in AI applications had shifted from "technical capability" to "engineering deployment and management."

"Big tech is building smarter soldiers, but no one's providing field manuals and command systems," Zhou Man realized. "If AI applications truly explode, 'how to manage AI labor' will become a more rigid demand than 'how to use AI.'"

ClawTroop was built to systematically solve the critical pain points of mass feeding, usage, scheduling, and management for enterprises and power users. Users can batch-create, deploy, and统一管理 their crayfish, assign them personalities, behavioral codes, and skills, and dispatch tasks to hundreds or thousands of agents at once while monitoring execution.

ClawTroop's interface looks almost defiantly plain. No slick 3D interactions, no real-time generated media streams — more like a legacy ERP admin dashboard: a neatly arranged "lobster list" on the left, four plain text input boxes on the right — Soul.md (personality), Agents.md (behavioral code), User.md (understanding of master), Tool.md (tool principles).

"We pulled out the core configuration items for crayfish so non-coders can easily adjust and modify them in text, and batch-apply to multiple agents," Zhou Man explained during a demo.

While the market chases drag-and-drop no-code interfaces, Moli Context insists users write configuration files in Markdown. The reasoning sounds almost techno-fundamentalist: Markdown is the most AI-interpretable structured language, and simultaneously the most natural thinking载体 for ordinary people.

But this design reflects deeper product insight: an AI agent's "personality" isn't parameter tuning, it's context construction. The four Markdown files form a complete cognitive framework — Soul.md answers "who are you," Agents.md answers "what can you do," User.md answers "who do you work for," Tool.md answers "how do you wield weapons." When users need to modify an agent's behavior, they don't hunt through nested menus for toggles — they add or delete a line of rules in text, click "batch push," and the entire cluster updates instantly.

Users can also group lobsters by tag — sales squad, customer service squad, Xiaohongshu squad. "Say you have 30 sales agents," Zhou Man offers as an example. "Previously you'd have to adjust scripts one by one. Now you just add one line to Agents.md — 'do not privately promise discounts exceeding 10%' — and the relevant lobster troops sync immediately. If any agent oversteps, the system flags it, and users can promptly add constraints and统一修正."

Part 02

Serving OPC: From Consumer to Business

This product, seemingly tailor-made for B2B SaaS, is targeting OPCs (One Person Companies) and consumer users in its first phase.

A self-media blogger is a one-person company. She can create five lobsters in ClawTroop to form a "Xiaohongshu Matrix Strike Team" — Agent A scrapes trending topics at 7 a.m., Agent B generates three titles and posts in different styles based on the topic, Agent C screens for banned terms and sensitive content, Agent D auto-publishes and monitors early metrics, Agent E plays "front-row commenter" in the comment section. The entire workflow requires zero code — just configure personality files in the backend and set behavioral rules like "if reads fall below 1,000, auto-adjust title keywords."

This pivot reveals the distinctive penetration path of AI applications: it may follow a reverse trajectory "from consumer enthusiasts to micro and small businesses, then upward," rather than traditional SaaS's "from large enterprises downward." When everyone can cheaply own an AI army, organizational boundaries, definitions of management, even the very concept of "work" itself will be rewritten.

Still, Moli Context faces inevitable questions: if OpenAI launches native Swarm multi-agent collaboration tomorrow with an out-of-the-box management interface, would ClawTroop be instantly obliterated?

A deeper paradox lies in the "cloud-raised crayfish" model itself, attempting to solve privacy anxiety — no local deployment needed, no frantic feeding of personal data, with security ensured through cloud container isolation and task-level context assignment. But this also means users must entrust core workflows to a third-party cloud.

ClawTroop's answer is "progressive intelligence" — an agent's smarts should be "grown through use," not "fed in advance." Rather than having users deploy locally first and frantically feed documents (raising privacy concerns with unstable results), it offers cloud-isolated containers where users assign context and file permissions per task. This eliminates privacy leakage risk while letting agents accumulate skills through task execution.

Yet this still doesn't evade the ultimate question: when agent scale expands from 10 to 1,000, does management complexity grow linearly or explode exponentially? Will the cost of solving these performance bottlenecks turn the promise of "low-cost AI armies" into a false proposition?

Zhou Man believes that competition at the AI application layer is ultimately competition in "productizing management capability," not mere "model capability." Big tech keeps delivering stronger models and infrastructure, but users still face concrete problems after acquiring these capabilities: what do I want my lobsters to do, how many do I need, how should I group them, what rules should each follow, how do I quickly fix problems when they arise — these aren't technical problems, they're management problems.

Moli Context will continue going deep into scenarios, onto the front lines of battle, exploring with OPC users how to deploy lobster troops in each specific business, and rapidly spreading mature experience through their community. This ability to embed with users, this community-building approach, is something big tech is structurally incapable of and holds no natural advantage in.

Layout by Du Meng | Images courtesy of company

Recommended Reading

The flow of money, the rise and fall of people