Summary of Key Points
This news article discusses the trend of AI Agent products (which can be understood as “AI assistants that can perform tasks on their own”) moving towards mobile devices. Tools like Claude Cowork, Cursor, and OpenClaw are no longer just chat windows on computers; they have evolved into “digital employee consoles” that can work in the background and be managed by users via their phones at any time. The significant change is that these tasks can run independently of the user's device (either in the cloud or locally), allowing users to focus on other tasks while only approving or making decisions at critical moments. However, there are still issues regarding stability, security of permissions, and cost.
Why Have AI Agents Suddenly Moved to Mobile Devices?
The reason isn’t for chatting; rather, they are being used as “remote assistants.” Previously, AI Agents were either available only on computers or served as chat interfaces on mobile devices. Things have changed because the user base has expanded: while they were mainly used by programmers for coding, now they are needed by operations personnel for organizing contracts, sales staff for creating client materials, and finance professionals for verifying expenses. These tasks often take dozens of minutes to hours and require coordination across multiple tools such as emails, calendars, and files. It’s not practical to be constantly glued to a computer.
For example, you can use Claude Cowork on your computer in the morning to assess the risks associated with renewing contracts for the third quarter, then attend a meeting. You can receive progress updates on your phone, and if there are uncertainties (such as whether to modify a clause), you can make a decision via the app, and it will continue working on its own. The phone has become more of a “remote control” for managing the AI’s tasks.
Three Products with Different Approaches to Mobile Deployment
Although all three products have moved to mobile platforms, their underlying operational logic varies:
- Claude Cowork and Cursor: Tasks are executed in the cloud. This means that even if you turn off your computer, the tasks continue to run. The cloud-based agents on these tools can install software and run tests, with results and merged code available on your phone. The advantage is a lower barrier to entry (no need for a constant online connection on your device), but the downside is that data may be stored on third-party servers.
- OpenClaw: Tasks still run locally on your computer, but you can control them remotely from your phone. You need to install a “gateway” on your computer first, and after pairing it with your phone, you can issue commands and monitor the status. The advantage is that all data remains under your control, ensuring security; however, your computer must be kept running, or the tasks will stop.
Users have mentioned that OpenClaw’s mobile app is more convenient for control because it allows direct connection to a private gateway, eliminating the need to share conversations with third-party platforms.
AI Agents Entering a New Phase: From “Real-Time Q&A” to “Self-Working Digital Employees”
This move towards mobile devices marks an evolution in AI Agent functionality:
1. First phase: Chatbots that answer your questions and complete tasks instantly, such as refining copywriting.
2. Second phase: Local assistants that can access files on your computer but stop working when the computer is turned off, requiring constant supervision.
3. Third phase: Asynchronous work agents that run independently of the device (in the cloud or locally). You provide the goals, and they handle the rest, only reaching out to you at critical points. The phone serves as an entry point for managing these tasks—checking task queues, statuses, and results.
The phrase “tasks can continue even when the computer is turned off” is crucial: previously, you might not have entrusted AI with tasks that required constant attention (like automatically generating a report at 6 a.m.), but now you can because you don’t need to be on your computer.
Challenges Faced by Current AI Agent Use
Despite their impressive capabilities, these tools are not yet fully mature:
- Stability issues: OpenClaw users have reported problems such as lost WhatsApp connections and unresponsive messages after upgrades, leading to interrupted workflows.
- Permission risks: As AI agents perform more tasks, they need access to your email, calendar, cloud storage, and even payment systems, which poses serious security risks if compromised.
- Cost concerns: Token-based pricing may result in high costs for long-running tasks, prompting users to seek fixed subscription plans.
- Reliability doubts: There are concerns about the accuracy of AI-generated results; for example, what if a contract is incorrectly organized? Therefore, manual approval processes are still necessary.
Some users have adopted a hybrid approach, using Claude to set up the tools and then having Cowork handle the tasks, with strict rules in place (e.g., requiring human confirmation for critical decisions), similar to hiring an employee who needs your approval before proceeding.
Future Competition Will Focus on “Backstage Infrastructure”
While model capabilities are important, what truly determines whether an AI Agent can be effectively used in work is the underlying infrastructure:
- Persistence: The ability to resume tasks after network disruptions or tool failures.
- Permission management: Deciding which actions (like sending emails or modifying files) require human approval and which can be automated.
- Transparency: Users need to understand what data the AI is processing, which tools it’s using, and why it stops at certain steps (to avoid being left in the dark).
- Cost control: Providing fixed subscription plans and budget limits to ensure users are aware of their expenses.
- Security measures: Protecting the AI’s data and access credentials from attacks.
Only by addressing these infrastructure issues can AI Agents become reliable “digital employees.” Otherwise, they remain mere decorations.
In summary, the move of AI Agents to mobile devices is not just a simple expansion of the client interface; it represents a significant upgrade in product functionality—from “you ask, I answer” to “you manage, I work.” However, there are many challenges that need to be overcome before they can become an integral part of daily work. The company that builds a solid infrastructure will have the upper hand in the future.