Returns triage, with the human in the loop you choose.

Paste a return case. An AI agent diagnoses it, recommends a resolution, decides whether it can act autonomously or needs human oversight, and drafts the customer message. The hard part of agentic products isn't the answer — it's knowing when the agent should act alone.

3 Operator decision

The agent recommends — the operator decides. In an autonomous case, this is a glance and a nod. In a human-in-the-loop case, this is where judgment lives. Either way, the agent did the legwork.

About this prototype

ReturnsAgent is a working prototype, built in an evening to make a point: the hard part of an agentic product isn't generating an answer — it's designing the right human–AI collaboration model for each case. When can the agent act alone? When does a human need to approve? This prototype makes that decision explicit on every case.

The triage is generated by a real LLM. It has no access to any company's actual order data, carrier integrations, fraud models, or returns policy — it reasons from general logistics knowledge. A production version would be grounded in real platform data, which is exactly where a platform with a billion shipments a year has a defensible advantage.

Built by Silas Kjærholm Mortensen as a conversation starter. The point isn't the tool — it's the cadence: hypothesis to working prototype in hours, not a quarter.