Human in the loop
Agents and tools · HITL, human approval
Human in the loop (HITL) means the agent does not finish on its own: at the important steps it waits for a person’s approval. Typical stopping points: granting permissions, writing to a live system, accepting the result. Next to it is human on the loop: the agent runs without stops, and a person watches its action log and can interrupt the work.
How much the person decides
| Human in the loop | Human on the loop | No human | |
|---|---|---|---|
| Who decides at an important step | the person approves | the agent; the person watches | the agent |
| When the person steps in | at every stopping point | when something looks wrong | never |
| Example for an agent | a prompt “allow writing to the database?” | the agent works, you watch the log and can stop it | an agent in a sandbox on staging |
| Where it fits | live systems, money, other people’s data | long tasks with known risks | anything easy to roll back |
Source: Wikipedia: Human-in-the-loop
Where it appears · 6
- AI agent = LLM + harness · Harness
- MCP server and AI agent for 1C · Harness
- Claude Code, Codex, Cursor and Hermes Agent · Harness
- Harness engineering in practice · Harness
- A reconcile loop for specs · Specs and context
More 1
- Prompts for work tasks · Prompts and review