Claude Code团队讲究啊，这都往外说

TL;DR - Anthropic’s Claude Code team says AI agents now handle 70%–80% of its routine work, shifting engineers from inspecting individual tool calls to defining goals and supervising persistent workflows. This matters because rapidly improving models are changing both software architecture and the engineer’s role toward higher-level problem solving.

• The team uses Slack-native Claude Tag to identify stakeholders, build prototypes, instrument and deploy tools, monitor feedback, and autonomously improve usage funnels.
• Claude Code is evolving from session-based assistance toward persistent, cloud-hosted agents that run background routines across multiple conversations.
• Rapid model improvements make capability-specific scaffolding short-lived, so the team favors composable primitives such as permissions, visualization, verification, code review, and feedback.
• Code review workflows use broad agent searches, adversarial checks, and multi-agent cross-validation to filter noise before escalating likely issues to humans.