Summary
Ben, a three-time founder running an AI agency and community for professionals, walks through seven prompting rules for Claude 5 models — specifically Claude Opus 5 and Claude Fable 5 — drawn from Anthropic’s official best-practices guides and internal team practices. The most significant shift he highlights: Claude 5 models perform best when handed the complete task specification upfront and left to run, rather than being guided step-by-step through a process. This represents a meaningful departure from how earlier Claude and GPT models were typically prompted.
The video directly quotes Boris Cherny’s keynote at Y Combinator, who frames the change as going “higher level” — describing the goal, guardrails, and exit criteria rather than issuing sequential instructions. Other rules covered include using an “interview me” skill to surface hidden requirements before starting a complex task, explaining the intent behind a request (not just the what), structuring prompts with markdown for complex multi-part tasks, and using voice memos as rough input that a structuring skill then polishes for the model.
Ben illustrates each rule with real examples from his own workflow: using Claude Fable 5 to pre-outline YouTube videos, planning a personal analytics dashboard with the interview-me approach, and building features inside his startup. The video also covers giving models explicit tool-use skills and the value of extended autonomous runs for complex builds — practical guidance for anyone working with Anthropic’s latest model generation.
📺 Source: Ben AI · Published September 02, 2026
🏷️ Format: Tutorial Demo
