Most chat products throw away their own best work.
An agent researches a topic, writes something useful, and then buries the result in a transcript. If you want to use it later, you copy it into a document, rename it, store it somewhere else, and lose the causal chain that produced it.
Rush takes the opposite approach.
What an artifact is
An artifact is the durable output of an agent run:
- a research report
- an email draft
- a table or comparison
- a checklist, plan, or brief
It is not residue. It is the unit of value.
Why the loop matters
The loop looks like this:
- an agent completes work
- Rush saves the output as an artifact
- you review or revise it
- the artifact gets shared, cited, or handed to the next agent
That loop is what turns one-off model output into compounding work.
What changes when artifacts persist
When the output stays inside the system, three things improve:
- continuity — tomorrow's work can start from yesterday's deliverable
- trust — you can inspect what the agent actually produced
- distribution — a shared artifact shows the product doing real work
A durable artifact does more than a polished answer in a chat bubble. It survives the conversation and becomes the next input.