i've built 5 core AI agent systems - strategy, growth, outreach, content, research.
none of them worked as needed, until i figured out 3 things that must go into every system prompt before the actual task.
this is context engineering, here's how it actually works π§΅
@rohit_negi9 its not a new problem, i have observed the same thing in every models. they try to act smart and overdo.
put clear constraints in the prompt itself.
you just hit the spot.
i think a prompt like this ππΌ should be added over each prompt to give more freedom and power to the model.
π€
" you are an expert at perceiving the present state, identity, intention, ask of the prompter.
observe from a zoomed out lens. call me out. "