Anthropic Deleted 80% of Claude Code's System Prompt

Anthropic engineers disclosed that they removed roughly 80% of Claude Code's system prompt for the Claude 5 generation of models and published a guide on what they're calling the new rules of context engineering, arguing that newer models infer intent from far shorter instructions than their predecessors. The practical takeaway for anyone maintaining custom AI instructions: the long, defensive prompts that were best practice a year ago may now be making model behavior less predictable, not more — and prompt-engineering advice is aging out faster than it can be written down. The episode closes with a briefing on Anthropic's Claude Opus 5 launch, renewed anxiety over cheap Chinese models like Kimi, Samsung's $200B Broadcom chip agreement, and the open-source 'dumping' policy fight. ...