Each Claude Code session starts fresh with only the system prompt and tools, but repo-wide context persists via a Claude.md (or Agents.md) file. Auto memory is on by default, letting Claude learn and carry forward context across interactions.
Anthropic's engineering blog details how internal code merges scaled from roughly 2,500 per day in May to over 30,000 per day by September, forcing them to build source control infrastructure designed for agent-scale rather than human-scale throughput. adocomplete's commentary: "Big problems need even bigger solutions."
Responding to skepticism about multi-agent persona setups, an Anthropic team member agreed that distinct agent personas are likely heading toward obsolescence as models become capable enough generalists to make role-play scaffolding unnecessary.
