Google 的高阶主管担忧公司在 AI 程式码工具领域落后，尤其是 Anthropic PBC 在企业市场快速渗透、吸引力更高。文章指出，Anthropic 与 OpenAI 正在吸引开发者，而 Google 的程式码能力仍然支离破碎：Gemini 的能力被分散在至少六项不同品牌的产品里，不少 Google 工程师甚至偏好使用 Claude Code。这被视为一个重大专注缺口。Google 已开始整合，首席 AI 架构师 Koray Kavukcuoglu 正在率领主要工程团队，将内部编码工作集中到名为 Antigravity 的平台；DeepMind 也透过由 Sebastian Borgeaud 领导的新团队扩大相关工作，John Jumper 亦参与其间。

在 Gemini 3 发表后，Google 曾被视为领先，但近月来动能转向实际可加速软体开发与除错流程的程式码助手。市场观察者认为，程式码是目前最直接且最具利润潜力的 AI 应用，因其直接影响交付速度与除错效率，因此创办人与开发者持续在 Claude Code 与 OpenAI Codex 之间切换以求最佳结果。Google 发言人指出其优势包含更强大的基础模型与巨大算力，而且 Alphabet 在二月表示公司内部约有 50% 的新程式码由 AI 辅助撰写，Antigravity 等工具据称已被广泛采用并加速模型与工具开发。

尽管有这些进展迹象，公司仍面临内部摩擦。许多团队只能透过商业案例申请例外才能使用竞争工具，且部分 DeepMind 团队据称依赖 Claude Code，因其实用。AI coding 的责任亦分散于 DeepMind、Google Cloud、Google Core、Google Labs 与 Android 之间，进一步造成不确定性。Google 以 2.4 billion 美元（约 24 亿美元）收购 Windsurf 的人才与技术来建置 Antigravity，但其组合仍拥挤，仍保有 Gemini Code Assist、Gemini CLI、AI Studio、Firebase Studio 及 Jules。瓶颈包括计算资源争用、效率与品质标准之间的文化张力，以及主管流动，因此观察者警告在快速移动的市场里，速度本身可能是唯一能持久的优势。

Google executives are worried the company is lagging in AI coding tools as rivals, especially Anthropic PBC, gain traction in business. The article says Anthropic and OpenAI are drawing developers while Google’s coding capabilities remain fragmented: Gemini features are split across at least six products with different branding, and even some Google engineers prefer Claude Code. This is seen as a major focus gap. Google has started consolidation anyway, with Chief AI Architect Koray Kavukcuoglu moving internal coding efforts toward a shared platform called Antigravity, while DeepMind is expanding work through a new team led by Sebastian Borgeaud and involving John Jumper.

Google was once seen as ahead after Gemini 3, but momentum shifted in recent months toward practical coding assistants that speed software development and debugging. According to market observers, coding is now the most directly lucrative AI use case because it directly affects delivery speed and debugging efficiency, so founders and developers keep switching between Claude Code and OpenAI Codex based on results. Google spokespersons cite strengths such as stronger foundation models and massive compute, and in February Alphabet said about 50% of newly written code is AI-assisted internally, with Antigravity and related tools reportedly adopted widely and accelerating model and tooling development.

The company still faces internal friction despite these signs of progress. Many teams can use competing tools only through business-case exceptions, and some DeepMind groups are said to rely on Claude Code because it is practical. Responsibility for AI coding is split across DeepMind, Google Cloud, Google Core, Google Labs and Android, adding to uncertainty. Google also acquired Windsurf’s talent and technology for 2.4 billion dollars (about 24 billion units? Actually 2.4 billion USD) for Antigravity, and its portfolio remains crowded with Gemini Code Assist, Gemini CLI, AI Studio, Firebase Studio, and Jules. Bottlenecks include compute contention, cultural tension between speed and quality standards, and executive turnover, so incumbents are warned that in a rapidly moving market, speed itself may be the only durable advantage.