3 ms·
You should make sure that your queries to the agent contain a "loop condition"[1]. This can be something like "keep trying until yarn test passes". Always reme
by skeptrune 10mo ago
You should make sure that your queries to the agent contain a "loop condition"[1]. This can be something like "keep trying until yarn test passes".
Always remember that these agents are just LLMs running tools in a loop and treat them as such.
[1] https://www.skeptrune.com/posts/prompting-the-agent-loop/ https://www.skeptrune.com/posts/prompting-the-agent-loop/
- verdverm 10mo agoif they aren't already, they will be multi-agent / llm systems, more and llm being asked to loop for example, in ADK you can set up a sequential agent with [planner, loop(coder, validator), finalizer] with the inner loop, such that another agent is the one checking and deciding if the task is complete or not, but we are also creating higher level enforced constructs of seq and loop Getting reliability in this high-level process is helpful