Write clear instructions Provide reference text Use external tools Split complex tasks into simpler subtasks Give the model time to "think" Test changes systema

Write clear instructions Provide reference text Use external tools Split complex tasks into simpler subtasks Give the model time to "think" Test changes systematically As these best practices solidify, why are they not being built into the UI or product itself for these tools? Seems trivially straightforward besides the last one. For the first one, add an optional persona field and allow query construction in pieces before sending over the wire. Permanently pre-prompt the model to always ask itself how long to "think" before answering, and ask itself intermediate questions if it's nontrivial.