Atrium: a tour of the AI investigation platform I run beside Snowflake CoCo

说明与用途

Atrium 是一个 Web 应用程序，包装 AI 编码代理的支持调查工作流，添加结构化问题陈述、来自不同 LLM 家族的对抗性审查和版本化反馈，以提高对代理输出的信任度。

问题

Support agents using AI coding agents produce confident answers without verifiable evidence, leading to mistrust and re‑verification effort.

结果

Atrium adds verifiable, adversarial review to AI‑assisted support, reducing re‑verification time and increasing trust in agent outputs.

工作流

• Pull cases from CRM mirror and display in case workspace with SLA timers
• Surface AI insights: similar past cases, likely root causes, SLA risk flags, suggested next steps
• Generate structured problem statement via LLM
• Agent investigates in terminal, Atrium listens to CLI hooks
• Run adversarial review (Mode 1: draft check, Mode 2: runtime rebuild) when clicked
• Inject versioned feedback packet into agent’s CLI
• Refine draft with tone options on dedicated screen
• Validate final reply in Slack and send to customer

证据

Atrium is a lightweight web app that runs right next to my coding agent's terminal session and wraps the whole life of a case: intake, SLA tracking, a structured problem statement, AI insights, an adversarial review of the agent's work by a model from a different family, and a final pass on the customer reply.

来源

The review has three design choices that matter more than any prompt: who reviews, what they check, and how the findings get back to the agent.

来源

Mode 1 tests the answer. It takes the customer-facing draft and cross-verifies the investigation behind it: it cross-references the case material and runs its own SQL queries, independently of the agent, to confirm or counter each point the draft makes.

来源

Mode 2 tests the path. It rebuilds the agent's run and checks it against three questions: Was the right flow used? Were key skills missed? Are there gaps or leaps in logic?

来源