Hugo Hahn

I recommend starting small with a single main agent using a reasoning model and tools. In the prompt add a "process" section which clearly states the steps it should take (step 1: orchestrate, step 2: review productboard etc...) For the custom instructions, add them in the main prompt. => I got pretty good results with it and it allowed me to iterate fast. Once you have a good understanding of your challenges, you can go the specialised agent root.

not resolved

Had the same issue, both with Claude 4 & OpenAI o3

As I was building an agent, iterating on the prompt, I would have loved to have 3-4 basic scenario where I expected a specific answer. It would allow me to iterate more peacefully, knowing I would not go back to a previous state.

Plus one on this.

Nope I did not, will do 🆗

the agent