Summary & Key Takeaways ​

• datasette-agent 0.3a0 has been released with new features.
• A new execute_write_sql tool allows AI agents to write to databases.
• This tool requires explicit user approval for all write operations.
• The chat terminal mode now supports these approval requests.
• New options include --unsafe for auto-approving actions and --root for elevated permissions.
• Tools can now provide plain text alternatives for CLI display.

Our Commentary ​

This is a smart move. Giving AI agents direct write access to databases always felt a bit... dicey. The user approval mechanism is a crucial guardrail. I'm also intrigued by the --unsafe mode; it's a clear acknowledgment of the tension between convenience and control. We're still figuring out how much autonomy to give these things.