When AI can produce a draft faster than you can explain what you want, the speed of communication starts to matter a lot.
There is an old idea behind this. In 1945, Vannevar Bush described the memex: a device for storing books, records, and communications so they could be consulted quickly. He imagined it as a supplement to human memory. You would not need to remember every detail if you could get back to the relevant record easily. That was a proposal for faster access to external knowledge, not a bigger biological memory. As We May Think
With AI, I think we have a similar opportunity in the other direction: getting an idea out of our heads and into something we can work on.
Until we can send our thoughts directly (some future day, maybe), we need to communicate them. A picture says a thousand words. Your opinion about that picture tells the agent which of those words matter.
You may struggle to describe the page you want from nothing. Show a rough version, and suddenly you can say: keep this layout, make that text larger, remove the background. The picture gives you something concrete to react to.
Then there are the questions you can answer before they arrive. Tim Ferriss, author of The 4-Hour Workweek, recommends including if/then instructions in emails to reduce back-and-forth. His essay on email overload gives the example of a shipment: if it has arrived, do this; if it has not, contact the person who can resolve it.
The same tactic makes sense with an AI agent:
Use this layout, but make the text larger. If the two columns become cramped on mobile, stack them. Keep all the information. If fitting it requires removing something, ask me first.
One message carries the reference, your judgment, and the next steps for likely situations.
You still need to check the result. And if you have not made a decision, the agent's question can be useful. But when you already know what you want in either case, tell it now.