FAQ: AI assistants, automation, and safety
FAQ: AI assistants, automation, and safety
Is AI required to use BusDK?
No. BusDK is CLI-first and deterministic without AI. AI-assisted flows are optional convenience layers on top of command and dataset contracts.
What should AI do in a BusDK workflow?
AI should help users compose, validate, and explain deterministic command flows. It should not replace core evidence, control gates, or explicit business decisions required by policy in workflow controls and compliance guidance.
Can AI run commands automatically?
AI-enabled interfaces such as bus-ledger and bus-factory can run repository-local commands, but approval and control behavior should remain explicit. Command execution should stay observable in logs and traceable to the active workflow context.
How should teams treat AI-generated output?
Treat AI output as draft guidance unless it has been verified by deterministic command results and domain controls. The authoritative result is always the reproducible workflow output produced by module commands such as bus-replay and bus-validate, not the model narrative.
How do we avoid leaking sensitive information to AI services?
Use strict input boundaries, explicit file selection, and secret reference handling. Teams should avoid sending unnecessary workspace material and should keep sensitive handling aligned with module-level security policy and bus-secrets.
Can AI-assisted workflows still be reproducible?
Yes, if AI is used to generate deterministic commands and those commands are replayed against the same declared inputs. Reproducibility depends on command/data contracts and deterministic .bus files, not on conversational text.
What should be logged in AI-enabled operations?
Logs should capture user actions, command intent, error details, and state transitions needed for diagnosis. High-signal logs make it possible to investigate behavior without relying on browser-only debugging, aligned with error handling and diagnostics and validation and safety checks.