Stop complaining about Claude's rate limits. Anthropic's Boris himself admitted: the pickiest users can't do without us. 别再骂 Claude 限速了，Anthropic Boris 亲口承认：最挑剔的用户，反而最离不开我们

The Rise of AI-Augmented Development: What Anthropic's Claude Code Reveals About the Future of Programming

The article presents a compelling snapshot of a paradigm shift in software development, as revealed through an interview with Boris, the head of Claude Code at Anthropic. Several key themes emerge that deserve deeper analysis.

---

1. A New Working Model: Developer as AI Orchestrator

The article opens with a striking observation: developers in Silicon Valley are adopting a dual role — acting as project managers during the day by assigning tasks to AI, and as cloud computing dispatchers at night by running hundreds of AI agents simultaneously.

This is not merely a trendy workflow adjustment. It signals a fundamental transformation in the nature of software development work:

• From writing code to directing code generation — The developer's core skill shifts from syntax mastery to problem decomposition and task orchestration.
• From individual contribution to system management — Running hundreds of agents means developers increasingly resemble operations engineers or system architects rather than traditional coders.
• From linear work to parallel processing — The ability to manage concurrent AI agents introduces a new dimension of productivity that was previously impossible.

This shift echoes historical transitions: just as assembly language programmers evolved into high-level language developers, today's developers are evolving into AI supervisors.

---

2. The "AI Writing AI" Loop: A Milestone with Nuance

Perhaps the most provocative claim in the article is that Claude Code is now 100% developed using Claude Code itself. Boris states this has been the case since the Opus 4.5 era.

This creates what the article describes as a "closed loop" of AI writing AI. Let's unpack why this matters:

• Practical validation — When a company uses its own product to build that very product (a practice known as dogfooding), it serves as the strongest possible endorsement of the tool's capability. Anthropic's willingness to bet its own development pipeline on Claude Code demonstrates genuine confidence.
• Accelerating improvement cycles — If AI improves the tool that improves AI, the rate of improvement can theoretically compound exponentially. Each improvement to Claude Code could accelerate the next improvement.
• Boris's important caveat — He emphasizes this is not yet fully automatic recursion. Humans still oversee, review, and guide the process. This honesty is significant — it distinguishes Anthropic's claim from hype and suggests a pragmatic approach to a genuinely revolutionary capability.

The deeper implication is that we are witnessing the early stages of recursive self-improvement, a concept long discussed in AI safety circles. Anthropic's internal acknowledgment that "the future will definitely move in this direction" shows they are both embracing this trajectory and presumably working to ensure it remains safe and controllable.

---

3. Explosive Growth: Numbers and Context

The growth metrics cited are staggering:

• Annual revenue: $4 billion → $45 billion (10x increase)
• Demand growth: 80-fold
• Claude Code becoming the entry point product for most Anthropic users
• Each model version update (Opus 4, Sonnet 4, Opus 4.5, 4.6, 4.7) driving exponential growth spikes

Several factors explain this trajectory:

• Network effects in developer communities — When influential developers adopt a tool, word spreads rapidly through technical communities, GitHub, and social media.
• Compounding capability improvements — Each model iteration makes Claude Code more capable, which attracts more users, which generates more feedback, which drives further improvements.
• Low barrier to entry — Boris describes Claude Code as a tool that lets people "build websites and software using everyday language." This democratization of programming opens the market far beyond professional developers.

The comparison to previous "viral" tech products is telling. Boris notes that even seasoned Silicon Valley veterans on his team have never seen growth like this, placing Claude Code in the company of historically transformative technologies.

---

4. The Token Limit Paradox: When AI Is "Too Productive"

An interesting practical challenge emerges: users are hitting token limits not because the AI is insufficient, but because it is too productive. The article describes token consumption growing "exponentially."

This reveals a fundamental tension:

• Current pricing models were designed for an era of limited AI interaction. When AI agents work continuously and autonomously, consumption patterns change dramatically.
• Heavy users — likely the most valuable and engaged customers — are the ones most affected, creating a potential retention risk.
• Boris's response (that most users don't frequently hit rate limits) is a classic median vs. tail distribution argument. While technically true, the power users driving the most innovation and value are precisely those in the affected tail.

This challenge will likely push Anthropic toward new pricing models, infrastructure scaling, or usage tiering strategies.

---

5. Product vs. API: A Dual Growth Strategy

Boris reveals that both Claude Code's direct product line and the API business are growing simultaneously, with the product side's share increasing significantly. This is strategically noteworthy:

• Early Anthropic was primarily an API company, providing models for others to build upon.
• Now, having direct consumer/developer products provides brand visibility, direct user feedback, and revenue diversification.
• The mention that building user-facing products also advances AI safety research is particularly interesting — it suggests that observing real-world AI usage patterns helps Anthropic understand and mitigate risks.

Boris's refusal to disclose which

这篇文章报道了AI编程领域的前沿实践与思考，核心信息清晰，可以从以下几个层面进行深入解读：

一、核心观点：AI编程的“自我实现”闭环

文章最引人瞩目的信息点在于 Claude Code已实现100%由自身开发。这不是一个简单的技术展示，而是一个具有里程碑意义的信号。

• “AI写AI”的闭环形成：这意味着AI不仅能辅助编程，更能进行自我迭代和改进。虽然负责人强调目前并非完全自动递归，但这个开端预示着软件工程的未来可能进入一个“自演进”的新时代。
• 工作模式的根本重构：文章开头描绘的“白天当项目经理，晚上当调度员”的场景，正是这种变革的直观体现。开发者角色从具体的代码编写者，转变为AI任务的规划者、协调者与监督者，人与AI的协作进入新的深度。

二、背景与现状：Claude Code的爆发式增长

Claude Code的成功并非偶然，而是技术积累与市场需求共振的结果。

• 产品定位精准：它超越了传统“聊天机器人”的范畴，通过强大的工具调用能力，实现了“用自然语言驱动软件构建”，直击开发者效率痛点。
• 增长数据惊人：文中提到的“年度需求暴涨80倍”、“预估营收从40亿攀升至450亿美元”等数据，直观反映了其市场统治力。每一次模型版本（Opus 4、4.5等）的迭代都引发了增长的“指数级跃迁”，证明了其持续的技术领先性。
• 战略意义凸显：对于Anthropic公司而言，Claude Code的崛起不仅带来了巨额收入，更成为其触达用户的首要入口，改变了以往过度依赖API业务的格局，实现了“自研产品”与“开放平台”的均衡发展。

三、逻辑解析：从工具到生态的演变

文章的深层逻辑链条是清晰的：

1. 技术突破驱动产品：模型在代码生成和工具调用能力上的精进（这是长期研发的核心方向），催生了Claude Code这款全新形态的产品。
2. 产品成功反哺技术：Claude Code的广泛应用产生了海量的、高质量的代码数据，这些数据又成为训练和优化下一代AI模型（包括Claude Code自身）的宝贵燃料，从而形成数据飞轮，加速了“AI写AI”的闭环进程。
3. 生态愿景逐渐清晰：Anthropic的策略是“两条腿走路”：一方面用Claude Code这样的明星产品树立标杆、占领市场；另一方面通过开放API和SDK，赋能万千企业构建自己的AI应用。这旨在构建一个以自家模型为核心的繁荣开发者生态。

四、深层含义与未来展望

这篇文章透露出的讯息，指向了几个值得关注的未来趋势：

• 编程本质的重新定义：编程正在从“手工艺”转变为“交响乐指挥”。核心技能将从语法记忆转向系统设计、需求拆解和AI管理。
• “AI自我改进”的加速：当AI能够高效地开发和优化自身时，技术迭代的速度可能会超出线性增长的预期。这既带来效率的飞跃，也引发了关于技术可控性与安全性的深层思考（这也是Anthropic公司一贯的关切点）。
• 商业竞争的维度变化：AI公司的竞争，已不仅是模型参数的竞争，更是工具链完整度、开发者体验和生态繁荣度的竞争。Claude Code的案例表明，一款顶尖的终端工具能成为整个业务增长的强劲引擎。

总结而言，这篇文章不仅是在报道一款明星产品的成功，更是在勾勒一幅由AI深度参与并重塑的未来软件开发图景。它标志着我们正从“人类使用AI工具”的时代，快步迈入 “人与AI协同进化，甚至由AI驱动AI进化” 的新阶段。