On the High Leverage podcast, host Joseph Ruscio sat down with Simon Willison to unpack the rise of AI‑assisted programming. Willison traced his own shift from championing “vibe coding”—a hands‑off approach where non‑programmers request code and accept the output—to what he now calls “agentic engineering,” where seasoned engineers leverage models like Claude Code to accelerate production‑grade work today in practice.

Initially he drew a hard line: vibe coding ignores code quality, suitable only for personal scripts, while agentic engineering demands security, maintainability, and performance. Recent experience, however, shows the boundary eroding—Willison admits he no longer reviews every line Claude Code generates, trusting automated tests and documentation. This mirrors real‑world practice of treating internal services as black boxes until issues arise.

The speed boost—jumping from a few hundred to thousands of lines per day—has upended traditional development pipelines, forcing teams to rethink design reviews and risk assessment. Willison concludes that AI tools amplify, rather than replace, expertise; seasoned engineers still steer direction, but the industry must accept new norms of trust and accountability when code emerges from models without a human reputation.