ByteDance's Doubao Client Transforms into Codex-style Agent with GUI and Cloud Computing Control
Read more
Pandaily
pandaily.com

ByteDance's Doubao Client Transforms into Codex-style Agent with GUI and Cloud Computing Control

ByteDance's Doubao PC client demonstrates high development speed, surpassing over 80% of startups. An analyst studying the product found that Doubao PC has effectively become a Codex-like agent, utilizing OpenAI's approach to capture the internal office market.

ChatGPT, once the world's largest general-purpose AI product, reached 1 billion monthly active users in May. In China, according to QuestMobile data, Doubao has 382 million monthly active users, representing a significant advantage among domestic AI-native applications.

OpenAI initially developed the autonomous Codex client and later integrated it into ChatGPT, transforming the chatbot into an executive agent. Doubao went further by embedding these capabilities directly into its PC version, rather than relying on a separate Codex module.

When testing a router purchase, the author sent two links from JD.com to Doubao. The client opened the login page in its built-in browser, retrieved full product information after authentication, and provided a comparative analysis.

Furthermore, Doubao supports remote control when the user is away from the computer. Since the phone and PC use the same account, the phone can send tasks to the computer, and users can select cloud computers, which the author believes OpenAI will soon copy.

In a complex GUI scenario, the author tasked Doubao with creating a survey on Tencent Surveys. After authorization, the system autonomously completed the survey creation, added questions, and edited answer options, controlling the graphical interface.

A similar mechanism is used when sending parcels via SF Express, filling out invoice forms, and comparing API prices between DeepSeek V4 Pro, GLM, Kimi, GPT 5.6 Sol, and Opus 5. Doubao functions as an agent with integrated connectors: the Feishu connector allows summarizing into Feishu documents, and the Tencent Meeting connector enables Doubao to work with meeting content.

For products lacking a Command Line Interface (CLI) or open interfaces, the GUI serves as a fallback option. The author warns about the risks of automated agent control on risk-sensitive platforms like Xiaohongshu.

The capabilities allow users to pin verified workflows: for instance, for analyzing video titles on YouTube for AI Maker Summit, the author created a 'Skill' that checks all confirmed presentation titles and provides detailed feedback in Chinese, with repeatable tasks being schedulable.

Seedance 2.5 comes bundled with Doubao membership quotas, including a built-in 'Skill' for video creation that structures requirements before generation. The author's principles include consciously determining if an agent can perform each task; abstracting repetitive work into 'Skills'; connecting agents to external systems, preferring connectors over CLI or MCP, using GUI as a last resort; and viewing AI usage as a skill honed over thousands of hours.

In conclusion, Doubao is a domestic counterpart to ChatGPT and a superior internal replacement for Codex.

Popular