Anthropic团队在两周的集中优化中，通过部署Claude AI模型进行性能监控与基准测试，使claude.ai网站和桌面应用的核心用户体验速度提升约3倍1。新页面加载时间从3.1秒降至0.55秒，Claude Code会话启动从0.8秒优化至0.3秒，Cowork云会话加载则从2.6秒改善到0.73秒1。这一优化预计每天可为用户节省数万小时的等待时间1。
团队通过在Slack频道中部署Claude进行实时监控，建立可测量的性能指标，使模型能够自主识别性能瓶颈并提出改进方案1。在整个优化周期内，团队合并了超过3000次代码变更，其中最忙的一天就部署了超过200次变更，且在整个过程中未出现任何面向客户的事件或代码回滚1。优化工作还引入了近200个功能标志以支持灵活部署，其中超过一半在冲刺期结束前被清理1。第三天内，团队已达成13个性能目标中的12个，前75个百分点的关键用户旅程也实现了目标达成1。
Anthropic accelerated the core user experience of claude.ai and its desktop application by approximately three times over a two-week period 1. The team deployed Claude AI models within a Slack channel to monitor performance, conduct benchmarking, and identify optimization opportunities 1. By establishing measurable performance metrics, Claude autonomously detected bottlenecks and proposed improvements, ultimately resulting in the merge of over 3,000 code changes without any customer-facing incidents or rollbacks 1.
The performance gains were substantial across multiple key workflows 1. New page loads improved from 3.1 seconds to 0.55 seconds, while launching Claude Code sessions decreased from 0.8 seconds to 0.3 seconds, and loading Cowork cloud sessions dropped from 2.6 seconds to 0.73 seconds 1. These reductions are estimated to save tens of thousands of hours of user wait time daily 1. The optimization effort achieved its targets for the top 75 percentile of critical user journeys, with the team reaching 12 of 13 goals by the third day of the sprint 1.
The sprint involved extensive code optimization work, with over 200 code changes deployed on the busiest day and approximately 150 optimization threads initiated, one of which generated nearly 60 pull requests 1. The team introduced close to 200 feature flags during the effort, with over half cleaned up by the sprint's conclusion 1. The optimization work was conducted using Claude Tag models in beta, configured to Opus 5.5 capability level 1.
评论
还没有评论，欢迎留下第一条。