How much should an AI decide without asking you first?

Zapier’s Sidekick AI workspace has iteratively redesigned its permission system to balance excessive approval prompts, which encourage users to approve reflexively, against unrestricted autonomy, which can undermine trust after an unexpected action. Its initial tool-based approach distinguished between read-only and modifying actions but proved inconsistent, difficult to manage across devices, and too vague for meaningful decisions. The team shifted to resource-based rules tied to files, apps, websites, and automations, with deterministic priorities in which denial overrides requests to ask, and asking overrides automatic approval. Subsequent changes made prompts clearer, moved reviews from chat messages into a dedicated panel and inbox, and enabled permissions to be expressed in plain language while preserving strict user-defined rules. Sidekick later introduced Auto mode, where an AI reviewer assesses actions not covered by rules, and YOLO mode, which allowed uncovered actions but retained protections for explicit restrictions and permission changes; pilot feedback and usability issues led the company to retire YOLO. A forthcoming mode will maintain one consistent rule set while using contextual judgments about reversibility and risk for uncovered actions, with built-in safeguards requiring approval for external communications and prohibiting legal agreements and payments.

No tracked trend matches for this post yet.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.