stopped writing 'dont touch prod' in my agents system prompt. wired the actual permission instead
Summary
Describes shifting from instructing an AI agent with a system prompt to avoid modifying production to implementing actual permission controls.
Similar Articles
Does anyone actually read the agent permission prompts anymore?
The article highlights the problem of users blindly clicking through permission prompts for coding agents, and presents a fork-based workspace approach (built in Gensee Crate) where agents work on isolated copies, enabling safe experimentation and parallel approaches, though external API calls remain a challenge.
Stop trying to prompt your way to agent safety. It's an access-control problem
The article argues that AI agent safety should be addressed as an access-control problem, recommending practices like least privilege, allow-lists, and approval gates to prevent unintended actions.
How do you stop coding agents from touching production data?
Discusses strategies to prevent AI coding agents from accidentally modifying production databases, advocating for read-only access, sandboxed environments, and approval gates over relying solely on prompts.
Those of you running AI agents in prod — how are you actually managing their permissions?
The article asks how engineers manage permissions for AI agents in production, highlighting common problems with broad access and lack of audit trails.
How much do you actually let an AI agent touch in production?
Discussion about scoping permissions for AI agents in production to avoid dangerous database actions, suggesting read-only mirrors, approval steps, or hard walls between suggestion and execution.