Making agents develop patience

API
,
1

So my goal is to develop agents that understands that they for example must wait for input, as in point 1 here:

I understand I can chain the agents deterministically with code, but I am trying to find the boundaries of its behavior. Specifically, I would like the agent to discuss patiently with the user, before handoffs/ tool calls etc.

A test agent is instructed to ask one question at the time, and wait for input. Just by running the agent in a while True loop, it asks a question every 10th second, i.e producing a final output, without corresponding input. Does anyone have experience “taming” the agents without the use of if/else, by evals, Temporal or anything else?

2

After experiments, the input() call in a while loop should generally block the execution, enforcing agent wait.

BTW I instructed the agent not to ask a question until the user said the keyword “OVER”. Immediately the agent started putting OVER at the end of every question :slight_smile:

1 Like