Human-in-the-loop AI design
Human-in-the-loop AI design gates consequential actions behind explicit user review. The human owns send, spend, and deploy; the model prepares the draft.
Start here
Patterns to ship first
Build playbook
Product examples
Frequently asked questions
What is human-in-the-loop AI design?
Human-in-the-loop (HITL) AI design requires a person to review and approve before the model executes consequential actions: send email, spend money, deploy code.
When is HITL required vs optional?
Required for irreversible or high-blast-radius actions. Optional or light for local, undoable edits on a creative canvas.
How is HITL different from an autonomy budget?
HITL gates a specific action with explicit review. Autonomy budgets cap how many steps or how long an agent runs unattended. Most agent products need both.
ChatGPT vs Gemini vs Perplexity HITL?
ChatGPT clarifies then handoffs. Gemini drafts in native app cards with a deliberate Send. Perplexity uses per-tool permissions before agents run.
Which HITL patterns should I spec first?
Start with suggest-confirm-execute for one irreversible action, then add granular consent if agents need standing tool access.