What changed
AI Studio users can set preconditions for tool inputs so tools run only when defined conditions are met and invalid inputs are not executed. Agentic Virtual Agent authors can also link virtual input variables to Knowledge tools even when the API does not require them, gating Knowledge tool execution on earlier tool outputs. The same release adds preset conversation hooks that inject turn-specific instructions when repetition is detected.
Vendor documentation (opens in a new tab)Why it matters
Business rules such as authenticate before answering account questions can be enforced in configuration instead of relying on prompt wording alone. That makes virtual agent behavior easier to test and to explain during review.
CxCave practical context
Who should care
- AI agent designers
- CX architects
Action
REVIEWMap the business steps that must precede each tool call, express them as preconditions or linked variables, and add regression tests for both the gated and ungated paths.
CxCave recommendation, not a vendor mandate unless explicitly stated.