Glossary · Workflow design
Human in the loop (HITL)
Also called: HITL, human review, human approval step
Human in the loop (HITL) is a design in which an automated process pauses at chosen points for a person to review, correct, approve or reject its output before it continues.
HITL isn't a lack of confidence in automation; it's deciding which decisions are worth a person's attention. Done well, people review a small share of items, the ones that are risky, unusual or irreversible, and the rest flows through.
Where to put a human
Score each step on two questions: how bad is a mistake? and how easy is it to undo? Put review where mistakes are costly and hard to reverse:
- messages sent to customers in your name,
- money: payments, refunds, discounts, invoices above a threshold,
- changes to important records (deal stages, account status),
- decisions about people (hiring, credit, access).
Keeping it fast
- Route, don't review everything. A router sends only flagged items to review.
- Show the evidence. Put the AI's reasoning and the source data next to the decision.
- Allow edits. "Approve with changes" beats rejecting and redoing.
- Set timeouts. Decide what happens when nobody answers: escalate, default to safe, or expire.
Regulators increasingly expect this for high-risk AI: the EU AI Act's Article 14 requires human oversight for high-risk systems. For everyday business automation, it's simply good practice.
In Fluxo Flow
The Human Review node pauses a run until someone approves, edits or rejects it, on every plan. Reviews land in the Review Queue with the prompt reviewers see, a priority and an optional assignee; reviewers can comment, act in bulk and filter by status, severity or assignee. A timeout sends unanswered items down a timeout branch, and Dry Run lets you simulate each outcome. Review alerts are shown in the app (there are no email alerts for reviews yet).
Read next
Related terms
- Approval workflowAn approval workflow is a process that routes a request (an invoice, an expense, a discount, a piece of content) to one or more designated people who must approve or reject it before it can proceed.
- AI guardrailsAI guardrails are automated checks placed before or after a model call that block, correct or route away inputs and outputs that break your rules, such as banned content, leaked data or invalid formats.
- AI agentAn AI agent is a system in which a language model decides its own next step, calls tools (APIs, searches, other systems), reads the results and repeats until it judges the goal reached.
- Dry runA dry run executes a workflow end to end with its side effects simulated, so you can see what every step would do without sending, writing or charging anything.
- Idempotency keyAn idempotency key is a unique value a client sends with a request so that, if the same request is retried, the server recognizes it and returns the original result instead of doing the work a second time.
- HMAC signatureAn HMAC signature is a hash of a message computed with a secret key that only the sender and receiver know, sent alongside the message so the receiver can prove it came from the sender and wasn't altered.