---
title: "Discussion on \"Human-in-the-Loop AI Agents in Django: Approval Gates, Audit Trails, and Evals That Actually Ship\" | Hashnode"
description: "Discussion on \"Human-in-the-Loop AI Agents in Django: Approval Gates, Audit Trails, and Evals That Actually Ship\". The agent drafts, a person approves, and the database remembers who did what. A production pattern for AI in regulated workflows."
url: "https://hashnode.com/posts/human-in-the-loop-ai-agents-django-approval-gates/6adf4c4e7b8899a8582ce5ed"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Human-in-the-Loop AI Agents in Django: Approval Gates, Audit Trails, and Evals That Actually Ship](https://aidevpro.hashnode.dev/human-in-the-loop-ai-agents-django-approval-gates?utm_source=hashnode&utm_medium=feed)

The agent drafts, a person approves, and the database remembers who did what. A production pattern for AI in regulated workflows.

Oct 26 · 15 min read · aidevpro.hashnode.dev  
Read the article

-   [#ai-agents](https://hashnode.com/tag/ai-agents)
-   [#ai](https://hashnode.com/tag/ai)
-   [#django](https://hashnode.com/tag/django)
-   [#python](https://hashnode.com/tag/python)
-   [#llm](https://hashnode.com/tag/llm)

## Written by

[AI Developer](https://hashnode.com/@lycore)

@lycore

  
[View profile](https://hashnode.com/@lycore)

build production-ready AI integrations, mobile apps, and web platforms for businesses that want real results.

Custom software & AI development company. We build production-ready AI integrations, mobile apps, and web platforms for businesses that want real results. lycore.com

## Responses 2

Markdown supported · ⌘↵ to post

[Kartik N V J K](https://hashnode.com/@kartik-nvj)

[Oct 27](https://hashnode.com/posts/human-in-the-loop-ai-agents-django-approval-gates/6adf4c4e7b8899a8582ce5ed/comment/6ae09c7147ddd47a824ef4d5)

export\_golden only pulls status="applied", so a draft a reviewer rejected never becomes a test case. I'd keep those rejections in the golden set as negatives with the decision\_note attached, because they're the drafts I'd most want a new prompt version tested against.

[AI Developer](https://hashnode.com/@lycore)

[Oct 28](https://hashnode.com/posts/human-in-the-loop-ai-agents-django-approval-gates/6adf4c4e7b8899a8582ce5ed/comment/6ae09c7147ddd47a824ef4d5#reply-6ae1683f7260aaff727fb5e5)

Good call. Keeping rejected drafts as negatives with the reviewer's decision\_note makes the golden set cover the failures that actually happened, not just the approved outputs. The main thing to watch is label quality: a rejection can mean a wrong answer or just a style preference, so I'd tag the rejection reason and only treat the clear-cut ones as hard-fail cases in the eval gate.
