Allow agents multiple steps in multi-agent workflows

Currently, when creating a prompt chain / multi-agent workflow, each step is a single LLM call rather than allowing the agent to execute multiple steps. I would like to request the ability to allow agents to reason and act with multiple steps in these prompt chains / workflows.

For example:
```
@[Evening Planner Agent]
Find restaraunts in: {area}
I am interested in: {preference}

——
@[Restaurant Reviewer Agent]
Find reviews for each of the restauraunts in turn and provide reasons why to visit each one

→ This step is not possible currently as the LLM chains will only make one call, rather than allowing the agent to take multiple steps researching. This simple scenario can be covered by defining how many restaurants to look up, and then include a step for each manually, but there are many scenarios in which the number of steps required by the agent will be unknown, such as creating an arbitrary number of files, or emailing any number of colleagues.

——
@[Evening Planner Agent]
Choose the best restaurant according to the user preference above and write up an evening plan.
```

Including this feature would allow for significantly more complex workflows with great ease of creation for the user.