Can't Afford to Keep Your AI Lobster? Zhou Hongyi Builds a Cloud Office with Personal Trainer

In 2026, the AI agent landscape is experiencing a curious divergence. On one side, developers are buzzing with excitement over frameworks like OpenClaw (nicknamed 'lobster'), HERMES, and OpenHuman, with rapid iterations and intense competition. On the other side, users are rapidly losing interest, with a wave of abandonment driven by frustration.

Many early adopters have gone from the thrill of successfully setting up a lobster to the headache of configuring environments, tuning skills, keeping computers running all day, burning through tokens, and constantly fixing workflows. The lobster itself is powerful, but it's genuinely hard to raise.

Acknowledging this, Zhou Hongyi, founder of 360, pointed out in a recent livestream that the three biggest problems are: too hard, too expensive, and unsafe. To address these, 360 has unveiled a secure cloud-based version of its AI agent, along with a 'Lobster Coach.'

The cloud version essentially provides a complete 'cloud office' for the lobster: cloud servers, cloud storage, cloud browser, etc. Users can give instructions from their phone, and the lobster executes tasks in the cloud. There's no need to keep a Mac running; tasks continue even when the device is off or disconnected.

The Lobster Coach takes over the most off-putting parts for ordinary users. Based on the task type, it guides users step-by-step to train a lobster, schedule agents, and improve workflows. In a sense, 360 is not just building another agent—it's building infrastructure for agents.

The article highlights real-world use cases. A design company owner used a lobster to automatically prioritize client feedback, summarize documents, and track preferences, freeing up mental energy. A marketing professional uses it to monitor industry news and compile competitive analyses, allowing them to focus on decision-making.

However, the biggest barrier remains that most lobsters feel like 'geek toys.' Ordinary users struggle to create standard operating procedures from past interactions or write effective prompts. The cloud version and coach aim to change that by providing a ready-to-use AI employee team.

In tests, the cloud version performed impressively. A user simply messaged the lobster app to create an industry report on AI video generation tools, and the lobster returned a comprehensive result with data and analysis—all without the user needing to keep their phone or computer active.

The Lobster Coach was equally effective. When asked to train a lobster to write articles in the style of a tech media outlet, the coach guided the user through defining characteristics, news judgment criteria, search scope, and working mode. Within minutes, a complete lobster architecture with soul, context, tools, memory, and skills was configured and deployed. Notably, the coach proactively added a task to scan AI news daily and included a rule to automatically correct writing methods based on data feedback.

360 has also pre-built thousands of expert lobsters for high-frequency scenarios, from image generation to market research. More complex tasks can be handled by multiple expert lobsters collaborating, each completing its part and handing off to the next.

Looking ahead, the article notes that while lobster's popularity has cooled, the underlying trend of agent adoption is accelerating. Gartner predicts that by the end of 2026, 40% of enterprise applications will incorporate task-based AI agents. The real challenge is not whether to use agents, but how to make them work in production environments.

Companies like Anthropic, Microsoft, and Google are all platformizing agent runtime, providing operations, security sandboxes, and auto-scaling. 360's cloud version aligns with this trend, packaging complex tasks so that non-technical users can also benefit from agents.

From installing lobsters for everyone and giving away tokens in March to launching the cloud version in May, 360 has been relentless. The core message is clear: the AI chat box era is about answering questions; the agent era will be about who can truly get work done for more people.