一位拥有15年经验的软件技术负责人指出,随着人工智能工具大幅降低上下文切换和知识获取的成本,许多过去被视为禁忌的软件工程做法如今已变得高效可行 1。该观点认为,行业传统的经验直觉建立在旧有的成本模型之上,而AI迅速改变了这一基础,导致基于过往规则的判断失效 1。因此,团队应当重新审视并调整那些旨在规避低效协作的传统管理策略 1。
在具体实践层面,部分开发团队正在实施新的合并请求流程:由Claude生成代码并开启PR,人类开发者作为所有者进行审查,若CI测试通过且Greptile审查未发现问题则直接合并,从而取消了传统的人工互审环节 1。此外,借助AI工具,团队成员可以通过询问Claude快速了解他人正在进行的并行工作流,这有效缓解了因拆分团队可能导致的知识孤岛担忧 1。除了代码审查,其他被重新评估的传统禁忌还包括从零重写代码、自建内部工具而非购买SaaS、多版本并行开发以及使用团队不熟悉的技术栈(如Rust或C#) 1。
作者进一步强调,工程师的核心价值正从掌握具体框架的深度知识转向架构判断力、产品思维和适应能力 1。鉴于AI能够即时提供特定技术的深度信息,职业建议已从推崇“T型人才”的领域专精度,转变为更重视工程基础、抽象设计能力以及对业务的理解 1。基于此,领导者被建议质疑基于旧成本模型的直觉,以适应这种由AI驱动的工作范式转变 1。
A software technical lead with 15 years of experience argues that artificial intelligence tools, specifically Claude and Greptile, are fundamentally disrupting traditional software engineering best practices 1. By significantly lowering the costs associated with context switching and knowledge acquisition, several approaches previously dismissed as "bad ideas"—such as eliminating peer code reviews and accepting knowledge silos—are now becoming efficient and viable strategies for modern development teams 1. This shift suggests that the core value of engineers is moving away from specific technical knowledge toward architectural judgment, product thinking, and adaptability 1.
The implementation of these new workflows involves distinct changes to standard operating procedures. One team has adopted a new pull request (PR) process where Claude generates the code and opens the PR, while a human owner reviews it; if the continuous integration (CI) passes and Greptile’s review finds no issues, the code is merged directly without traditional human-to-human peer review 1. Furthermore, concerns about knowledge silos resulting from splitting teams have been mitigated because members can quickly query Claude to understand parallel workflows undertaken by colleagues, thereby reducing the friction of context switching 1.
These adjustments reflect a broader change in the underlying cost structure of the industry. Past rules were built on assumptions about the high time cost of manual reviews and the difficulty of acquiring knowledge, but AI has altered these economic realities, rendering judgments based on old intuitions obsolete 1. Consequently, professional advice is shifting from promoting "T-shaped" deep expertise in specific domains to emphasizing foundational engineering skills, abstract design capabilities, product mindset, and business understanding, as AI can provide instant depth in specific frameworks 1. Other traditionally taboo practices currently being reassessed include rewriting code from scratch, building internal tools instead of purchasing SaaS products, maintaining multiple version branches, and utilizing technology stacks unfamiliar to the team, such as Rust or C# 1.
评论
还没有评论,欢迎留下第一条。