The first useful Rush moment is not the install. It is the first time you ask an agent to do real work and get back something you can actually use.
1. Install an agent
Pick a specialist. For example:
rush install prix/deep-researcher
That pulls the agent from the registry and makes it available locally.
2. Run the task
Now dispatch it on a prompt:
rush run prix/deep-researcher --prompt "Research the AI note-taking market and summarize the strongest product gaps."
Rush also supports local agent paths when you are developing:
rush run ./agents/my-agent --prompt "test"
3. Look for the artifact, not the transcript
A strong run should leave behind usable output: a report, a draft, a comparison, a checklist. That is the point of the system.
In Rush, the conversation starts the work. It is not the work. The output should be something you can hand to another person or another agent without asking them to reconstruct the whole thread.
4. Iterate by delegation
Once the first agent returns something useful, the next step is not more copy-paste. It is follow-on work:
- refine the output
- hand it to another specialist
- turn the artifact into the next task
That is the real shift. You are no longer operating one tool at a time. You are directing work through agents.