glossary · Human decisions

Human-in-the-loop

definition

Human-in-the-loop is a design pattern where an automated process pauses at defined points for a human decision, rather than the human supervising continuously or not at all.

in one line: The process stops at named moments for a person to decide.

in context

Where does human-in-the-loop fit?

Stop exactly where it matters. In ConvOps, the work stops only at the moments you choose: the ones where a wrong move is expensive.

fix the bugthe AI
run the teststhe AI
releaseapproval
shippedafter your OK
two ways to say it

What is human-in-the-loop, in plain words?

Same idea at two depths: the plain version, then what the engine actually does.

Watching everything does not scale. Trusting everything is not governance. So you name the moments that need a person, and the work stops there.

Add checkpoints where a person, a metric or a report should decide. Leave them out where the work can run by itself.

for engineers

The requires_approval gate holds the step until an explicit approval is given, recorded with the account that gave it.

Checkpoints are set per step, so adding one or taking one away is an edit to the workflow, not a change to the agents.

the longer answer

Why does human-in-the-loop matter?

The pattern's value is precision: instead of choosing between watching everything (which does not scale) and trusting everything (which is not governance), you name the exact moments where human judgment is required and make the system stop there.

Implementation quality varies enormously. Asking the model to "wait for approval" is etiquette, not a mechanism. A real implementation makes the pause structural: the process cannot continue until the approval is given.

The pattern also lets you choose your checkpoints. Add them where a person, a metric or a report should decide. Leave them out where the work can run by itself. A checkpoint is part of the process, so moving one is an edit to the process, not a change to the agents.

It helps to separate human-in-the-loop from its neighbours. Human-on-the-loop means a person monitors a running system and can step in, but the system does not wait. Human-out-of-the-loop means full automation. Human-in-the-loop is the only one of the three where the system is designed to stop and wait, which is why it suits irreversible actions: shipping code, moving money, contacting a customer. The choice is less about trust in the model than about the cost of being wrong: the more expensive and the less reversible the action, the stronger the case for a stop.

in convops

How does human-in-the-loop work in ConvOps?

In ConvOps, the in-the-loop moment is a step with a requires_approval gate. The gate holds the step until an explicit approval is given, recorded with the account that gave it, and a require_context gate can demand that the reason for a decision is recorded before the step closes. Because every step is persisted, the person does not need to be in the same session: the run waits at the gate, and the approval can come later, from another session. Adding or removing a gate is an edit to that one step.

questions

What do people ask about human-in-the-loop?

What is human-in-the-loop in ConvOps?

Human-in-the-loop is a design where an automated process stops at defined points for a human decision. In ConvOps that point is a step with a requires_approval gate, which holds the step until an explicit approval is given, recorded with the account that gave it. Because every step is persisted, the approval can come later, from another session or another client.

What is the difference between human-in-the-loop and human-on-the-loop?

In the loop, the system stops and waits for a person at a defined point; on the loop, a person watches a running system and can step in, but nothing waits. ConvOps builds the first with gates on workflow steps. Human-in-the-loop suits irreversible actions such as shipping code or moving money, and human-on-the-loop suits low-risk, high-volume work.

How do you add human-in-the-loop approval to an AI agent?

Make the pause structural rather than a request in the prompt. In ConvOps, put a requires_approval gate on the step: it holds the step until an explicit approval is given, recorded with the account that gave it, from any session. If the step names an on_rejected target, a rejection sends the run to that step.

Does the EU AI Act require human-in-the-loop?

Not as one fixed pattern: Article 14 requires that high-risk AI systems can be effectively overseen, including the ability to intervene or stop the system. ConvOps approval gates are one way to make that oversight concrete and recorded, alongside runs that can be stopped and an audit of each advance. Whether your system is high-risk is a question for counsel.

Which roles act as the human in the loop?

In ConvOps it is whoever owns the decision at that point: a reviewer for a merge, an editor for a publish, a finance lead for a payment, a support lead for a customer reply. That person gives the approval from any connected client, recorded with the account that gave it. ConvOps does not reserve approvals for people today: any caller with access to the step, a run included, can give one, and the record shows which account did.