Summary of Key Points
This article discusses the transformation in the working methods of the Claude Code team, part of the Anthropic AI company, a native AI organization. In the past, software engineering processes revolved around the high cost of writing code; however, in the era of AI, writing code is no longer the bottleneck. The challenges have shifted to areas such as verification, review, and security. As a result, the team has restructured their planning methods, automation habits, code review processes, and role divisions, actively eliminating inefficient procedures to allow humans to focus on creative judgment and critical decision-making, while leaving AI to handle repetitive tasks and execute details.
#### 1. Planning Methods: From “Drawing Blueprints in Advance” to “Building Prototypes for Testing”
In the past, projects required lengthy design documents and multiple review meetings due to the high cost of coding and concerns about errors. The Claude team realized that with AI, which writes code extremely fast, a six-month roadmap can become outdated in just three months. They now adopt an “Just-In-Time (JIT)” approach:
- Eliminating Unnecessary Rituals: Instead of creating complex design documents, they directly build prototypes (e.g., using AI to quickly generate functional demos) for internal users to use immediately.
- Early Verification: Since the types of bugs in AI-generated code have changed, they have moved testing and feedback processes forward, allowing for immediate modifications while the system is being used.
- Prototyping Instead of Arguing: If two team members disagree on a plan, they ask AI to create two prototypes within ten minutes. This approach is ten times more efficient than arguing over PowerPoint presentations.
In simple terms, instead of drawing 100 pages of blueprints before building a house, they first build a small model house, live in it, and make adjustments immediately if issues arise, avoiding unnecessary effort.
#### 2. Automation: Use AI for Repetitive Tasks to Free Up Time
Whenever the Claude team encounters repetitive work, they ask themselves, “Can this be automated?”
- Starting with Small Steps: For example, Fiona used to manually summarize customer feedback every day; now, she uses AI to automate this task, so she can focus on other tasks while drinking coffee.
- Reversing the Cost-Effectiveness Ratio: Previously, automation required writing complex scripts and was only worthwhile for high-frequency, critical tasks. With AI, the cost of automation is almost zero for tasks that are repeated more than three times.
- Building a Foundation Gradually: There’s no need to establish an entire automation system from scratch; handling one small task each day can significantly change your work approach over time.
It’s like manually organizing expense reports every day; now, AI can automatically identify invoices and fill out forms, freeing up time for more valuable tasks.
#### 3. Code Review: Let AI Handle the Mundane Work, Humans Make Critical Judgments
With AI generating code so quickly, how can humans keep up with the review process? The Claude team’s approach is to “trust but verify”:
- AI Handles the Basic Work: Tasks such as style checking, bug detection, and additional testing, which account for 60-70% of the review workload, are handled by AI.
- Humans Handle Critical Decisions: Legal compliance, security-sensitive code (e.g., handling user data), and product quality (e.g., ensuring features meet user needs) are areas where AI cannot replace human expertise.
- Dynamic Boundary Adjustment: As AI models improve, tasks that previously required human intervention may be automated, requiring continuous re-evaluation of what still needs to be done manually.
This is similar to teachers grading assignments: AI handles mechanical tasks like multiple-choice and fill-in-the-blank questions, while teachers focus on more creative and logical assessments like essay writing, doubling efficiency.
#### 4. Team Roles: Blurred Boundaries in Recruitment
AI has blurred the boundaries between different functions within the team:
- Mixed Roles: Project managers (PMs) can write code, and engineers can create design documents. For example, after fixing a bug, AI generates user copywriting, which is then revised by humans before release.
- New Recruitment Criteria: Instead of evaluating how many lines of code someone can write per hour, the team looks for two key abilities:
1. Creative Builders: Those who know what needs to be done and can quickly use AI to create prototypes (creativity is valuable; typing speed is not).
2. System Experts: Those who can identify subtle errors that AI may miss (e.g., hidden risks in code logic).
In other words, while AI can perform tasks, humans are responsible for deciding what to do and how well it’s done—this is where “taste” or judgment comes into play.
#### 5. Driving Change: Actively Eliminating Inefficient Processes to Maintain Flat Hierarchies
In the AI era, old processes can become mere formalities. The Claude team takes proactive steps:
- Flat Management: Managers start by working on the front lines and support flexible team mobility (e.g., assigning people to teams based on their expertise).
- Prioritize AI: Tasks that can be automated are done by AI to free up resources for more challenging work.
- Proactive Process Elimination: If a process, like weekly meetings, is not effective, they simply question its necessity and cancel it.
Many companies’ outdated processes (e.g., useless weekly meetings or redundant approvals) are maintained out of inertia; someone needs to step forward and say, “This can be stopped.”
#### Conclusion: AI-Native Organizations Are About More Than Just Buying Tools
The exploration by the Claude team demonstrates that an AI-native organization is not about simply acquiring AI tools for employees. It’s about rethinking everything, from planning and review to recruitment. Even top teams are still figuring out the best practices (e.g., whether to have separate iOS/Android development teams or where to draw the line for fully automated reviews). The core principle remains the same: use AI to replace repetitive tasks and allow humans to focus on judgment and creativity.
As Fiona puts it, “Identify the most cumbersome parts of your work process and ask yourself—does it still deserve its current place?”
This analysis explains the impact of AI on organizations in plain language, helping you understand that AI is not just a tool but also a catalyst for changing work methods.