"Agents" become the apps of the AI era.

Lei Jun's "Tiger Brother" has returned.

Recently, a founding team that has participated in the development of the last three generations of operating systems has attracted much attention. It includes Hugo Barra (nicknamed "Tiger Brother"), former VP of Product Management at Google Android and former Global VP at Xiaomi, David Singleton, former VP of Product Engineering at Android, and Ficus Kirkpatrick, former VP at Meta (responsible for VR and AR platform development).

Tiger Brother, Image/Xiaomi

The reason for this "dream team" to come together is not hard to guess - AI. Specifically, they have jointly founded a startup, /dev/agents, with the goal of developing the next-generation operating system for AI agents based on the cloud, making AI agents more than just isolated tools and building a "smart ecosystem" that can seamlessly collaborate across devices.

It's not just /dev/agents planning to reinvent the operating system. Companies ranging from Google, Apple, Microsoft, OpenAI, to Huawei, Honor, vivo, OPPO, and more have already begun their layouts and actions.

This is a step destined to be taken. From personal computers to the internet and the proliferation of mobile devices, almost every revolution in computer technology has followed a similar path - hardware breakthroughs leading to software prosperity, with the operating system defining the new rules at the core.

It's not surprising. Without an operating system, hardware is just a pile of silicon chips, and programs are just rigid binary instructions. This is the key reason why the operating system is considered the "soul" of a computer: it connects hardware and software, developers, and user experiences through a common language.

In today's world where AI is consuming everything, a new operating system is needed.

AI operating system, a must-fight battlefield for giants

A few days ago, Huawei launched the Huawei Mate 70 series at the Mate brand festival, with AI unsurprisingly becoming one of the series' biggest highlights. Building on the system-level AI agent Xiaoyi, the Huawei Mate 70 series introduced nine AI functions.

But ultimately, the AI in the Huawei Mate 70 series cannot be separated from another protagonist of this launch event - HarmonyOS NEXT (native HarmonyOS).

Image/Leitech

In the battlefield of AI operating systems, there are many participants. Globally, players entering the field can be roughly divided into two categories: large model vendors represented by OpenAI, Anthropic, and Doubao, and mobile phone and PC vendors ranging from Google to Huawei, from Honor to Lenovo.

These two forces enter through distinct paths but converge on the same goal - dominating the next era of operating systems.

Large model vendors enter with a cool and rational approach, and their logic is clear: since they possess the most powerful AI models and can provide more "intelligently emerging" AI agents, they naturally have the opportunity to create more native AI operating systems and terminals that better adapt to the times. In fact, large model vendors including OpenAI, Anthropic, and Doubao are indeed integrating and transforming various terminals.

Claude's "Computer Use", Zhipu's AutoGLM agent, and even OpenAI's "Operator" expected to be launched in January next year are all enabling AI to operate autonomously on mainstream computing terminals such as mobile phones and PCs, attempting to bypass traditional interaction forms and directly replace user operations to reconnect hardware with software, developers, and users.

Image/Claude

In contrast, mobile phone and PC vendors adopt a more pragmatic approach, generally reshaping the operating system experience with AI as the core based on existing major computing platforms. From Google's initial integration of Gemini in Android 15 to Honor's system-level AI in MagicOS 9.0 - YOYO agent, Apple, Huawei, vivo, OPPO, and Lenovo are no exception.

"Agents" become the apps of the AI era

Every technological wave changes the interaction logic of devices. In the PC era, Windows liberated the keyboard and mouse, making computers mainstream production tools; in the mobile internet era, Android and iOS turned smartphones into the center of life through touch and the APP ecosystem.

Today, AI is driving a new interaction revolution, with the AI operating system serving as the underlying support for this change. But what exactly is an AI operating system? The answer to this question seems to be becoming clearer with technological iterations and explorations.

Essentially, iterations and upgrades to operating systems are all aimed at further improving the efficiency and experience of human-computer communication and collaboration. Unlike mobile operating systems that use touch as the means and applications as the core for human-computer interaction, under the latest wave of large model technological changes, AI operating systems enable human-computer interaction through agents using multimodal intelligent interaction.

In general, AI operating systems are attempting to transform devices from "passive response" to "active perception", making devices intelligent partners rather than tools. This may sound complicated, but the logic behind it is straightforward: users no longer need to remember how to use devices; instead, devices need to understand how to serve users.

For example, when you say to a system-level AI agent, "Help me organize an XXX meeting," after understanding your intent, the AI operating system breaks it down into tasks such as generating meeting invitations, sending them to relevant colleagues, reserving meeting rooms, and organizing meeting materials, and intelligently schedules sub-agents, third-party agents, and apps for autonomous operation.

Zhao Ming ordering 2,000 drinks through YOYO, Image/Leitech

In fact, this intelligent capability has already been initially demonstrated in systems such as Honor Magic 9.0 and Huawei HarmonyOS NEXT. Honor's MagicOS 9.0 is a typical example, where at the launch event, Honor CEO Zhao Ming directly ordered "2,000 drinks with one sentence" through the YOYO agent.

Behind this is not only the scheduling of underlying hardware and software but also the AI operating system's active understanding of user intent based on user device data, the breakdown and execution of tasks, and the intelligent scheduling of first-party or third-party agents and apps. This represents a new challenge in the era of different apps.

Image/Leitech

Two "writing styles" of AI operating systems

If AI operating systems are the core pieces in the next technological revolution, their implementation paths are not unique. In today's terminal battlefield, AI operating systems are presenting two distinct writing styles:

One is the "reformist" approach represented by Apple and Google, which deeply integrates AI capabilities into the existing operating system architecture to create AI-enabled operating systems.

The other is the "native" approach represented by vivo BlueOS and Huawei HarmonyOS NEXT, which redesigns the architecture from the ground up with AI at its core.

The advantage of the reformist approach lies in its "robustness." For example, Apple has proposed the concept of a personalized operating system, integrating the AI experience across the entire ecosystem through Apple Intelligence, seamlessly connecting AI functions with users' existing habits while maintaining the original app ecosystem.

Image/Apple

For instance, in Apple's iOS and macOS systems, traditional operating systems are transformed through a series of incremental AI functions. Especially in the upcoming iOS 18, AI is further deeply integrated. From Siri's intelligent upgrades to AI-based health tracking and data organization, Apple's AI system is ubiquitous yet always hidden beneath user habits, smooth and restrained.

However, it also has obvious bottlenecks. The architecture of traditional operating systems is not designed for AI tasks and is often constrained by performance and logic when facing large-scale parallel computing or complex scene perception.

In contrast, vivo BlueOS and Huawei HarmonyOS have chosen a more radical path - redesigning the underlying logic of the operating system with AI as the core for native architecture design. Taking vivo BlueOS as an example, it incorporates large model technology from the outset and even uses Rust, which Musk considers the "best language for achieving AGI," as the underlying language for system development.

Image/vivo

Simultaneously, in terms of architectural design, BlueOS emphasizes AI capabilities at the system level, which is actually similar to native HarmonyOS. Both are "intent-centric," with mobile phones and other terminal devices serving as tools to execute user intent rather than mere application carriers. In the native intelligent architecture of HarmonyOS, the intent framework and user data graph are also designed as a subsystem.

Image/Huawei

The advantage of the native approach lies in its "future potential," enabling it to fully unleash the potential of AI technology and adapt to more complex scenarios in the future. However, it also faces significant challenges. Designing from scratch means that the developer ecosystem needs to be rebuilt, making it difficult to quickly implement and generate benefits in the short term. Even native HarmonyOS has taken a decade and still faces considerable challenges.

"Reform" or "native"? It doesn't matter

A few months ago, Eric Schmidt, former CEO of Google, gave a speech at Stanford University, specifically mentioning that current AI technology is like early electric technology during the Second Industrial Revolution, with everyone simply "reaching out to pick the low-hanging fruit." AI operating systems are breaking through this situation, applying AI capabilities more broadly and deeply.

But which will ultimately prevail - the reformist or native approach? The answer to this question may not be simple. The challenge for the native approach lies in the difficulty of ecosystem construction, which inevitably requires a longer time to mature. Native HarmonyOS is no exception. However, it is clear that native operating systems in the AI era have no historical burdens and greater potential.

At least for now, the key to AI operating systems may not lie in whether they are "native" or "reformed" but in who can faster and more accurately understand the relationship and interaction between humans and machines under multimodal intelligent interaction. It is foreseeable that for some time to come, these two models will likely continue to evolve and iterate towards the same general direction.

Regardless of how the path evolves, the race for AI operating systems has only just begun, and the winners will ultimately be those vendors who can make "devices understand people better."

   Source: Leitech