Skip to content

Human-in-the-loop

πŸ“ Short definition

Principle that an agent or automated system may analyze and propose, but must wait for human validation before a critical action: merge, publication, deletion, access to sensitive data.

πŸ“– Detailed definition

In an agentic workflow, the LLM can read files, call tools, produce a patch, or open a PR. Human-in-the-loop places a human as the control point before the irreversible step.

This differs from human-on-the-loop, where a human supervises after the fact. For a document vault or Git repo, the recommended model is hybrid: the agent may work on a branch, but merge stays human.

πŸ’‘ Why it matters for on-prem AI

On-prem does not guarantee safety by itself: a local agent can still delete, publish, or over-edit. Human-in-the-loop limits operational risk while keeping automation gains.

⚠️ Common pitfalls

  • Confusing notification with validation: receiving a report is not approving the action.
  • Allowing automatic merge to main for editorial tasks.
  • Letting an agent read archived plans and execute stale TODOs.

πŸ“š Go deeper

  1. Human-in-the-loop workflow
  2. Vision: custodian agent

πŸ”— See also

Give my opinion

Everything is optional: one answer is enough. This site sends nothing: you send it, the way you choose.

Your noteNo note
Your opinion is about
About what? (several choices possible)
What will be sent
Send by e-mail

Private, no account: opens your mail program with the message ready, for ia-on-prem@damien.becherini.fr.

Open a GitHub issue β†—

Public, with a GitHub account. You read the issue again before you publish it.

To paste it wherever you want.

Give at least a note, a subject or a message to be able to send.