ByteDance’s Doubao 2.1 Pro Shifts Focus to Developers: Coding, Agents, and the B2B Turn in AI

For developers like Lin Ming, the shift from using Cursor to tools like Claude Code and Codex marks more than just a change in software—it’s a transformation in how AI integrates into the development process. Once seen as a "co-pilot" that assisted with small code snippets, AI now plays a central role in R&D workflows: models like Codex can split tasks, read code, call tools, run tests, and deliver complete solutions. Codex’s rapid growth underscores this trend: its weekly active users have surged over 7 times in six months, exceeding 5 million as reported.

ByteDance is now joining this shift with its latest Doubao 2.1 Pro, which adds coding and agent capabilities. During its launch, Volcano Engine repeatedly compared the model to industry leaders like Claude Opus, indicating that ByteDance is no longer just chasing consumer user numbers but targeting developers, agent applications, and real-world workstreams. This pivot is a significant departure for a company long known for its consumer-focused products and traffic-driven growth.

The industry context for this shift is clear. The 2025 Stack Overflow Developer Survey shows that 84% of respondents use or plan to use AI tools in their development process, up from 76% the previous year, with nearly half (47.1%) using AI coding tools daily. Gartner predicts that by 2028, 90% of enterprise software engineers will use AI code assistants—up from less than 14% in early 2024.

Cost pressures are also driving this pivot. For ByteDance, Doubao’s high computing power costs (reportedly millions of yuan daily) contrast sharply with its low daily revenue (under 1 million yuan, mostly from e-commerce commissions). Internally, ByteDance has reallocated resources to enterprise services and coding models, expanding its data review team from 1,500 to 3,000 people to clean data for coding models. The company’s TRAE team, which uses AI to write 90% of its code, has seen only a 60% increase in per-person task throughput—highlighting the need to turn AI-generated code into deliverable, production-ready outputs.

Competitors are also making moves in the AI coding space. Zhipu’s GLM5.2 model has coding capabilities approaching Claude Opus but at one-seventh the price. Anthropic’s Claude Code, launched in May last year, has already reached an annualized revenue of 25 billion USD by February 2026, with over half coming from enterprise clients—including 1,000+ clients spending over 1 million USD annually, up from 500+ two months prior.

This shift toward B2B AI signals a key realization: free consumer chat apps may generate buzz, but they struggle to sustain profitability. For ByteDance and others, focusing on developers and enterprise workflows—where AI tools can deliver measurable value and predictable revenue—is becoming the path forward.

Sources

• Ebrun (mobile site): https://m.ebrun.com/685575.html
• Stack Overflow 2025 Developer Survey
• Gartner
• LatePost