Rich Hickey Recasts LLMs as Automation

Hickey’s strongest point is about the human cost of delegation, though calling LLMs “automation” can obscure interactive uses where developers retain meaningful control.

• –The tool-versus-automation distinction highlights how coding agents can replace practice instead of extending skill.
• –Summarization and generated code may remove the learning that comes from reading, debugging, and forming arguments directly.
• –LLM unpredictability makes feedback loops, verification, and bounded workflows essential for developer tools.
• –The debate suggests choosing AI interaction models based on the goal: use assistance for learning, automation for execution, and rigorous review for both.