ChatGPT has made 'Text inputs with no affordances' a popular talking point. While users learn over time what works, the burden of discovering effective prompts

ChatGPT has made 'Text inputs with no affordances' a popular talking point. While users learn over time what works, the burden of discovering effective prompts remains. How can we improve this experience?