Show HN: Current – Change control for AI agents (Elixir) (github.com)

Current, a new tool for managing AI agents developed in Elixir, introduces a robust change control framework that enhances safety and accountability in AI-driven development. Unlike existing models like Claude Code or Codex, which allow AI agents to act autonomously, Current operates on a proposal basis. Agents can only suggest modifications while awaiting human approval, ensuring that all changes are validated and documented in a tamper-evident log. This approach addresses concerns regarding the unpredictable nature of AI behaviors, making the integration of AI in software projects more secure and manageable. The significant implications for the AI/ML community include enhanced reproducibility and verifiability. Current's architecture confines agents' actions to a predetermined set of intents and restricts their access to only necessary project files. This system creates a reliable audit trail, enabling developers to trace the origin and authorization of every change. Each proposal undergoes a thorough validation process, requiring approval before any alterations to the project are applied, which helps maintain project integrity. By enabling robust oversight and documentation, Current positions itself as a vital tool for organizations seeking to leverage AI safely and effectively in their development workflows.