ByteDance appears to resume smartphone R&D based on recruitment data.

By Sleepy and Siwei Guaiguai

In December 2025, the long-anticipated "Doubao Phone" finally debuted. It integrated the Doubao Phone Assistant technical preview into the Nubia M153 engineering prototype, with an initial price of 3,499 yuan; approximately 30,000 units sold out on the day of launch.

In the days shortly after its release, its price at the seafood market once soared several times higher than its original value. The Beating editorial team even purchased two units.

It’s not because it’s an exceptionally useful phone—in fact, as a “technology preview,” the first-generation DouBao phone offered a less-than-ideal experience. What excites us most is that, for the first time, it pulled AI out of the chat box, transforming it from a chatbot into an AI agent capable of controlling a smartphone.

On the DouBao phone, the AI can see your screen, understand what you're browsing, hear your voice, switch between different apps, and directly assist you with many tasks—such as checking train tickets, comparing prices, claiming coupons and placing orders, or editing photos. Although sensitive actions like payments still require your personal confirmation, the AI is already capable of independently completing many tasks that previously required you to manually click through each step.

Although it’s still a bit clunky, sometimes slow to respond, and occasionally freezes—like someone just learning to use a smartphone—it truly gave us our first intuitive sense of how convenient AI can be in everyday life.

Later, the lobster emerged and became a global sensation; AI Agents became another iPhone moment in AI following the release of ChatGPT, prompting numerous companies and entrepreneurs to sell computers and phones preloaded with OpenClaw. DouBao Phone was at least one version ahead of them and can even be considered a pioneer of this AI Agent wave.

Unfortunately, DouBao Mobile soon faced a crackdown from major tech companies. Access or functionality was gradually blocked on platforms such as WeChat, Taobao, Alipay, and banking apps. Some called it a "ban," while others claimed it was merely a trigger of risk control—but for users, the difference was negligible: it simply stopped working.

We are very sorry. DouBao Phone is certainly not a mature consumer electronics product, but it has shown the entire industry the雏形 of the next-generation gateway.

Even though the hype around DouBao Phone has passed, we haven’t fully let go of the matter. Recently, our routine data collection captured thousands of job postings, and upon analysis, it appears that ByteDance may be restarting its smartphone development.

We extracted three categories from ByteDance's official recruitment page: AI Innovation Business, Mobile OS, and DouBao Mobile Assistant.

After deduplicating by job ID, we further extracted detailed page information and cross-referenced keywords from the job title, job description, and requirements.

Unlike typical AI app team hiring, ByteDance’s recent batch of experienced hires includes roles in mobile operating systems, cameras, touch control, connectivity, battery life, heat management, chip integration, structural design, manufacturing processes, and production line testing.

These terms are not common in internet companies; they are things that phone manufacturers, supply chain companies, and engineering teams deal with daily.

ByteDance is hiring people to work in factories.

However, this does not confirm that ByteDance will launch its own smartphone brand, but it does indicate that they are restarting development of mobile-grade devices.

Now let’s look at what these positions themselves reveal.

First, look at the DouBao mobile assistant.

We conducted a more focused screening, searching the original data for job postings containing the term "Doubao Mobile Assistant" in the title, description, or requirements, resulting in a total of 83 positions. These positions can be categorized into three distinct types, which together form the shape of a system-level AI Agent.

Category one roles are responsible for enabling AI to possess Agent capabilities.

For example, the position "Agent Development Engineer - DouBao Mobile Assistant" requires the AI to perform task decomposition, context organization, tool invocation, memory retrieval, state management, result validation, and exception recovery—these are the foundational capabilities of all current AI agents.

Category two roles are responsible for giving AI agents a good memory.

The roles have emerged in areas such as "perception and memory," "user memory," "personal knowledge graph," and "long-term preferences." For an AI agent to truly integrate into our lives, it cannot treat us as strangers each day—it needs reliable, stable long-term memory.

Of course, this easily raises privacy and boundary concerns, but according to the job postings, ByteDance has at least begun developing "memory" as one of the most important capabilities of the DouBao mobile assistant.

The third category of roles is responsible for enabling AI agents to deliver those capabilities on mobile devices.

For DouBao Mobile Assistant to operate a user’s phone on their behalf, it cannot exist solely in the cloud or function as just an app. It requires a comprehensive set of capabilities—including models, memory, task execution, on-device deployment, system applications, audio and video processing, communication, testing, and quality assurance—to understand user speech, interpret its environment, coordinate across devices, remain ready at all times, and operate reliably without errors.

Now let’s look at mobile OS.

There are 236 positions related to mobile OS, primarily located in Beijing, Shanghai, and Shenzhen. The most frequently recurring terms in the job descriptions are kernel, chip, driver, camera, display, audio, network, power consumption, thermal management, and mass production delivery—terms that are all closely tied to hardware and the underlying mobile system.

For example, the job description for the "Kernel Lead - Mobile OS" role states that the position requires leading the memory and storage team to adapt and develop the kernel for new Qualcomm platforms, ensuring seamless compatibility with mainstream smartphone chips and effectively managing memory and storage on mobile devices. These capabilities are critical for an AI agent to achieve real-time responsiveness and handle tasks in the background.

For example, the job description also includes terms like SoC, BSP, and RTOS. SoC can be roughly understood as the core chip of a smartphone, BSP is a set of low-level software that enables the system and hardware to recognize and work together, and RTOS is commonly used in scenarios with high demands on responsiveness and power consumption.

Therefore, the hiring for mobile OS roles signals that ByteDance is seeking professionals who understand mobile-grade terminal systems—individuals who know where AI agents running on smartphones encounter permission issues, power consumption bottlenecks, and system stability challenges, and which problems require collaboration with chipmakers, manufacturers, and testing teams.

Judging from the job requirements currently being advertised, ByteDance has already entered the deep waters of the smartphone industry.

Additionally, it is worth highlighting separately the positions located in Shenzhen.

Positions based in Shenzhen are often related to hardware, supply chain, testing, and mass production, whereas those in Beijing tend to focus more on models, algorithms, and platforms, and those in Shanghai lean more toward products and engineering.

If a project is only a cloud service, Shenzhen isn’t that important; but once it involves physical products, Shenzhen becomes crucial.

What we’ve seen in the relevant positions in Shenzhen are precisely these things.

Some positions are listed as human-computer interaction design, covering physical device interaction, software interface interaction, and cross-device coordinated experiences. These roles don’t just focus on designing interfaces within screens—they also consider the tactile feel of physical devices, button design, wake-up mechanisms, and how devices work together seamlessly.

Some roles are closer to the engineering site, such as interconnect, power consumption, short-range communication, baseband, system manufacturing, structural design, and test manufacturing.

These terms may not sound as impressive as "agent," "multimodal," or "world model," but in consumer electronics, it’s ultimately these things that determine success or failure.

If ByteDance merely wanted to make DouBao a better mobile app, it wouldn’t need to go through all this trouble. Once it started hiring for these roles, it signaled that it was already ready to commit to this path.

In the past, phones were containers for apps; in the AI era, they may become the bodies of agents.

If a phone is merely a container for apps, companies like ByteDance can build their empires through content, algorithms, and product strength, one app at a time. But if the phone becomes the body of an agent, and users first issue tasks, whoever takes on those tasks gets to determine the subsequent path.

In this path, the app will be downgraded to a callable tool. This will make all super apps uncomfortable, because agents are inherently designed to bypass intermediate layers.

So, the real challenge may not be whether Doubao can open an app, but whether others are willing to let it do so. An AI that makes decisions on behalf of users cannot be easily approved like a regular app.

For an agent to move from the chat interface to the operational layer, they must take on a wide range of tedious, messy tasks that previously fell outside the AI team’s responsibilities. They need to understand when the system will terminate background processes, which operations trigger risk controls, why phones overheat, and why factory yield rates remain low. These issues once weren’t the AI team’s concern—but now, they can’t be avoided.

So ByteDance is hiring for these positions. It may not actually launch a smartphone, but ByteDance can no longer afford to be just an app inside other people’s phones.

For large model companies to become the next-generation user gateway, they cannot remain dependent on others' operating systems forever.