Multimodality, hardware, and coding are among the most prominent artificial intelligence trends shaping the latter half of 2024. The first two areas may be familiar to many, given the strides made by various industry players. However, AI coding—virtually unheard of last year—has swiftly gained traction, emerging as a new narrative in the application layer of AI.
The latest development in this space is the resignation of Chen Zhijie, TikTok’s algorithm lead, who is reportedly planning to launch his own venture centered on AI coding.
Before joining ByteDance, Chen served as chief technical architect at Baidu from 2010 to 2019, where he was primarily responsible for advertising algorithms. After joining ByteDance in 2020, he led algorithm development for domestic platforms like Douyin, Toutiao, and Xigua Video. By 2022, he had risen to the role of TikTok’s algorithm lead, overseeing recommendation systems and data science teams to support video, live streaming, e-commerce, and content distribution.
2024 witnessed the rise of AI coding products such as Devin, Cursor, and OpenAI o1, which garnered global traction and renewed attention for this niche within AI applications. Baidu CEO Robin Li spotlighted AI coding as a promising avenue during the company’s Q3 director-level meeting, while Alibaba Cloud and ByteDance introduced their own offerings—Tongyi Lingma and MarsCode, respectively.
“Coding stands out due to its low susceptibility to ‘hallucination’ compared to long-text generation,” an algorithm engineer at Moonshot AI told 36Kr. This reliability has driven commercial use cases, such as Kimi’s assisted programming, which received internal approval.
Startups are also flourishing. Su Wen, a former investor at China Growth Capital, founded AIGCode in January 2024 and secured two funding rounds. Meanwhile, Ming Chaoping, previously a product lead at Moonshot AI, launched Xinyan Yima, valued at USD 50 million in its angel funding round—comparable to early valuations of leading Chinese AI unicorns.
Despite varied strategies, many AI coding startups aim to go global. “Coding is free from cultural or linguistic constraints, unlike language-based applications, which require significant regional adjustments,” an investor told 36Kr. “AI coding products are inherently standardizable, reducing barriers to scaling.”
The path to success in AI coding hinges on two factors: model capabilities and customer willingness to pay. Overseas markets, where top coding models like OpenAI’s o1 and Claude excel in handling complex natural language inputs, offer fertile ground. In contrast, many domestic models struggle with long-sequence processing and require auxiliary systems like retrieval-augmented generation (RAG) for stability.
Claude’s integration into StackBlitz’s Bolt illustrates this potential. Within four weeks, Bolt achieved USD 4 million in annual recurring revenue (ARR) and attracted over 100,000 weekly users. “Claude 3.5 Sonnet is the enabling technology that made this product possible,” said Eric Simons, Bolt’s co-founder and CEO.
The surge in AI coding startups and investments stems from its position as one of the few monetizable AI applications today. With algorithm engineers indispensable across industries, the productivity gains enabled by large models in coding are undeniable. A 2024 survey by Menlo Ventures revealed that AI coding products achieved the highest adoption rate among major AI application scenarios, reaching 51%.
This growing demand has transformed AI coding into a multimillion-dollar revenue generator. In December 2024, The Information reported a tenfold increase in annualized revenue for Anthropic’s software development and code generation business over just three months. Similarly, Microsoft disclosed in its July financial report that GitHub Copilot’s annual recurring revenue (ARR) hit USD 300 million, contributing 40% of GitHub’s overall revenue growth.
Cursor, however, has showcased even steeper growth. Research firm Sacra estimated Cursor’s ARR in November 2024 at USD 65 million, marking a staggering year-on-year increase of 6,400%.
Aspiring to replicate such growth trajectories, many AI coding startups are entering a market recognized for its vast opportunities, numerous niche scenarios, and relatively low competition. Yet, some industry professionals argue that the ultimate differentiator for these startups will not be their model sophistication but their ability to productize.
Currently, most players aim to emulate Cursor by developing copilot-style products that facilitate natural language dialogue. The success of these tools hinges on their seamless integration with integrated development environments (IDEs) and their capacity to blend smoothly into developers’ workflows.
“Productization is a core strength of many Chinese AI startups with visibility overseas,” an industry expert told 36Kr. “This is also why investors see significant potential for Chinese AI coding ventures to succeed globally.”
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Zhou Xinyu for 36Kr.