Claude Code团队讲究啊，这都往外说

Claude Code's team now delegates 70-80% of daily work to Slack-native AI agent Claude Tag, shifting from reviewing every tool call to assigning goals and trusting the model - exemplified by an internal tool built via Claude Tag that handles stakeholder discovery, prototyping, instrumentation, and funnel optimization autonomously. They maintain deliberate "non-attachment" to features, removing once-critical scaffolding like the to-do list (obsolete after Sonnet 3.5's memory gains) and AskUserQuestion (superseded by direct HTML artifact generation), instead composing replaceable "primitives" (Permissions, Visualization, Verification) as model capabilities advance every two months. The team raises abstraction layers from tokens to sessions to goals to persistent background routines, using Claude-written workflows for adversarial multi-angle code review, while reaffirming that engineering fundamentally remains problem-solving.

evidence

4 sources carry this story · confidence medium

topic

AI Tools & Agent Workflows

source

量子位