How to automate buyer replies with Unknown
The engineer reports 1,467+ replies automatically sent to real buyers and unattended operation of a chat channel without an official API.
news.ycombinator.comPrompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Build a RAG support agent for [seller] from [approved chat corpus]. Distill Q&A, define two confidence gates, evaluate in shadow mode, then limit automatic replies to [whitelisted scope]. Report results and unresolved session-management requirements.
Reach people exploring AI agents
See placements →What happened
The engineer reports 1,467+ replies automatically sent to real buyers and unattended operation of a chat channel without an official API.
Reported by leoycwan on Hacker NewsHow the agent did it
- Distilled roughly 70,000 chats into 8,400 Q&A pairs.
- Added two fail-safe confidence gates.
- Ran the customer-service agent in shadow mode first.
Where a person stepped in
The author used confidence gates, shadow mode and a whitelist; the excerpt does not describe individual human reply review.
A person set the goal and reviewed the result.
An engineer described shipping a retrieval-augmented customer-service agent for an online seller. The source material for its knowledge included roughly 70,000 real chats, distilled into 8,400 question-and-answer pairs. The account focuses on deployment safeguards as well as the eventual automatic replies.
The engineer added two fail-safe confidence gates, ran the system in shadow mode first and then moved to a whitelisted rollout. That sequence is the reported path from preparation to sending replies to real buyers. The excerpt does not give the thresholds, retrieval design or criteria for the whitelist, so those details cannot be reconstructed from the post.
The engineer reported more than 1,467 replies sent automatically and said the chat channel ran unattended. The channel lacked an official API and its session expired roughly every 24 hours; the post reports unattended operation despite those constraints without explaining the session-handling mechanism. Although Claude API appears in the author's technology list, the agent's underlying model is not identified. Who it's for: customer-service teams considering cautious automation built around existing conversations and staged deployment.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · leoycwan
Frequently asked questions
How was the support knowledge prepared?
The engineer reports distilling about 70,000 real chats into 8,400 question-and-answer pairs.
What safeguards preceded automatic replies?
The post names two fail-safe confidence gates, shadow mode first and a subsequent whitelisted rollout, without specifying their thresholds.
More details
Remaining steps
- Rolled out to a whitelist and sent automatic buyer replies.
About this account
The deployed customer-service agent sent buyer replies automatically and the author reports unattended chat-channel operation.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=50030810. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: leoycwan on Hacker News
Source published 10 Oct 2026
Source checked 14 Oct 2026
Tools and topics
Custom RAG agent · Q&A corpus · Confidence gates
Customer support · RAG · Staged rollout
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.