Hacker News 上的一篇文章指出，作者回顾了从使用 Open Router 等工具转向 Anthropic 模型和 Claude Code 的过程，并批评近期的 Claude Opus 5 模型与 Claude Code 工具更新变得冗长且过度工程化 1。该作者认为，早期的 Claude 模型以简洁和聚焦著称，但近期的变化导致用户需要“与工具对抗”才能完成工作 1。针对 Opus 5 被评价为“懒惰、草率和冗长”，且在排名中落后于 gpt-5.6-sol、luna、kimi k3 及 deepseek v4 flash 的反馈，Anthropic 发布了 Concise Output Style 以应对其冗长输出问题 1。
在工具迭代方面，Claude Code 的 /doctor 命令从原本的运行设置与确认功能，被重构为对所有提示词和系统的全面审计 1。此外，Anthropic 近期还发布了 AI-native SDLC Playbook 1。文章作者将上述问题归因于模型本身、工具迭代节奏以及营销方向，并呼吁 Anthropic 控制“AI 膨胀”（AI Bloat）、回归基础，以免竞争优势被竞争对手蚕食 1。为对抗“AI 膨胀”，作者提出了五个框架问题，分别是“做还是不做”、“现在还是以后”、“这么多还是少一点”、“这种方式还是更简单的方式”以及“自己做还是交给别人” 1。
A recent discussion highlights growing user frustration with Anthropic's latest releases, as an author reflects on their transition from tools like Open Router to Anthropic models and Claude Code 1. While early Claude models were praised for being concise and focused, the recent updates to Claude Opus 5 and Claude Code have been criticized for becoming overly verbose and over-engineered, forcing users to fight against the tools just to get work done 1. User feedback has described Opus 5 as lazy, sloppy, and verbose, noting that it ranks lower than competitors such as gpt-5.6-sol, luna, kimi k3, and deepseek v4 flash 1.
Specific changes to the tooling have also drawn scrutiny, particularly the restructuring of the /doctor command in Claude Code, which shifted from simply running setups and confirmations to auditing all prompts and systems 1. In response to the complaints regarding Opus 5's lengthy outputs, Anthropic has introduced a Concise Output Style 1. Additionally, the company recently released an AI-native SDLC Playbook 1.
Attributing these issues to the model itself, the pace of tool iteration, and marketing directions, the author urges Anthropic to return to basics and curb AI bloat before its competitive advantages are eroded by rivals 1. To combat this bloat, the author proposes a framework of five key questions: To do or not to do, Now or later, This much or less, This way or easier, and Ourselves or someone else 1.
评论
还没有评论，欢迎留下第一条。