Is Doubao’s AI Phone the Future iPhone?

Doubao recently released a technical preview of its AI‑powered smartphone. The author performed a series of everyday tasks to assess its capabilities, focusing on screen‑recognition‑based automation, latency, privacy implications, and overall user experience.

Advantages

The screen‑recognition approach bypasses closed‑source app restrictions, enabling actions such as sending WeChat messages, posting to Moments, and adding custom remarks (e.g., adding “no ice” to a latte order). This method also allows limited control over WeChat, which is otherwise inaccessible due to revoked APIs.

Disadvantages

Each UI transition requires a fresh screen analysis, resulting in noticeable delay; a complete ordering process took six minutes. The system also struggles with privacy, continuously listening and potentially capturing unintended speech, and it can perform high‑risk operations like transfers without user confirmation.

Test Scenarios

Cross‑app command (AI Agent intelligence)

Command: “Order a Luckin coconut latte on Meituan, half‑sugar, no ice, deliver to the office.” Result: The process lasted six minutes, slower than manual ordering. Doubao correctly added the “no ice” remark and auto‑selected a coupon, but failed to click complex “free expansion” red‑packet actions. Honor’s YOYO completed a similar order in one minute.

Nearby high‑rated restaurant navigation

Command: “Find a restaurant with a rating above 4.8 nearby and navigate there.” Result: Doubao launched Gaode navigation but used its own limited database for restaurant search, yielding distant and incomplete results. Honor’s YOYO leveraged Baidu Maps for a faster, more comprehensive search.

Driving time to airport and ride‑hailing

Command: “Check how long it takes to drive to the airport and book a car if traffic is clear.” Result: Doubao queried the map app for the route, then opened a ride‑hailing page for user selection—more steps than YOYO, which directly opened Gaode’s ride‑hailing interface, making it quicker.

WeChat payment

Command: “Send a 200 CNY red packet to [contact] with the note ‘buy something tasty.’” Result: Doubao filled the amount and note automatically; the final payment required manual confirmation. The flow worked well.

WeChat messaging

Command: “Message Yichen on WeChat ‘Good evening.’” Result: Doubao opened WeChat, searched by pinyin, asked for clarification when multiple contacts matched, and sent a slightly verbose greeting. Tencent has since blocked this behavior, indicating a risk of account suspension.

Address navigation from a chat

Command: “Navigate to the address sent in the chat.” Result: Doubao used screen‑recognition to extract the address and launched a navigation app, but occasionally failed to detect the address. Honor’s OCR was more accurate.

Package tracking

Command: “Check the status of this courier number.” Result: Doubao attempted an online query, failed, and provided a web link for manual lookup. Honor’s YOYO directly called the Cainiao API and displayed results in a card.

Conclusion

Only a few manufacturers, such as OPPO and Honor, focus on AI‑centric phones. Compared with Magic7, which offers faster AI interactions, Doubao’s screen‑recognition method is innovative but suffers from high latency, privacy concerns, and susceptibility to platform countermeasures. Future adoption will likely require deeper hardware integration and cooperation with app ecosystems to balance control and user privacy.