Put It Right in Front of Me.
When your agent needs your OK, the thing you're approving belongs right beside the question. Never “see above.”
- What my agent wrote“shown above”
- What I asked for“right in front of me”
- What it does nowThe item. Then the question.
“Shown above.”
One Saturday in October I was approving a stack of short paragraphs my agent had written for a project, one after another. Each one needed my OK before it went in. In between, updates kept rolling in about everything else we had running. Then this landed:
The restaurant paragraph is shown above: eight real, long-established places, each verified open in 2025–26. It waits for your approval.
↑ “Shown above.” Somewhere up there, in the scroll. Six seconds later:
In the future when you ask me to approve something and add this to the notes for future sessions too because you guys all do this. When you ask me to approve something, put what you want me to approve right in front of me at that time because I'm spending so much time trying to find anything.
Understood. Saving that as a standing rule for every future session, then putting the restaurant paragraph right in front of you.
[…] The rule: whatever I ask you to approve sits right next to the question, never “shown above”.
Here's the one waiting now, the restaurant paragraph […]
And then the whole paragraph, right there in the message, with the question under it. I approved it twenty seconds later. Over the next four minutes I approved four pieces of text that way.
Six feet up the scroll.
Here's what “shown above” looks like from my chair. I work with a lot of agents, and every one of them talks. Status updates, test results, “done,” “still running.” When one asks me to approve something “above,” I've got two choices. I can go hunting for it, or I can approve something I haven't read.
The first wastes my time. The second is worse, because an approval of something you didn't read isn't an approval at all. It's a guess with your name on it, and Lesson 8 was all about guesses.
Ten days later I was in the middle of fixing something when an agent slipped a short to-do list for my morning into the same conversation. I said:
“by the time this is done it's gonna be 6 feet up the scroll”
My agent answered, “I'll repost your morning list fresh at the bottom, so you won't need to scroll for it.” A couple of minutes later, when the fix was done, there it was: “As promised, here's your morning list:”. That's the whole idea. The thing I need goes where I'm looking, at the moment I need it.
What “right in front of me” looks like.
In my studio this is a written rule now, and it has three parts.
The whole thing, not a pointer to it.
The exact text, the numbers, the list, the picture, in the same message as the question and the last thing I read. Not “see above,” not “the draft I shared earlier,” not a file name. If it was shown before, show it again.
One question at a time.
Back in September an agent kept sending me messages with two, three, even four yes-or-no questions in each. I told it: “lots of questions.. ask me one at at time”. (Yes, “at at.”) When I fire back a quick yes, yes, ok to a pile of questions, nobody can be sure which yes went with which question, including me.
The recommendation first.
Tell me what you'd do, then the details. One morning I got pages of hardware prices and no answer. I said: “I did see one tag that said recommended, but it was in the middle of a lot of hardware. So there was no summary recommendation.” My agent put a one-page recommendation in front of the prices, “so the answer comes before any prices.” Most of the time I just say yes.
Sixty-five questions? Make it a form.
One at a time works until the questions pile up. One evening in November my agents had dozens of them waiting on me. One at a time, that's an evening of back-and-forth. So I asked:
“On the 64 open questions, could you give me a form I could fill out and hit submit on, and then you would have all those answers?”
So my agent built one. Every question became a card: the question in plain words, my agent's recommendation, a couple of alternatives and a box for my own words. By the time it was built it held 65. I answered 64 of them in one sitting and hit submit. On 48, the recommendation was right and I took it. On 11, I wrote my own answer. On 5, I picked one of the alternatives.
One card asked me to choose between two versions of the music for one of my games, and I didn't remember ever hearing either one. So in that card's box I wrote:
“please include with each question anything that I might need to audition because I don't remember ever hearing a score for […], and that's likely because it passed me in the thousands of words that I ignore every day unintentionally”
That's now the rule for every form my agents build me: whatever I need to see or hear to decide, the clip or the picture, goes right on the card. The next night, just after midnight, I asked my agent to make sure every future agent would know how to do it:
“You know how the other day I had like 65 questions and you did it in a form with everything I needed right there? Is that in our Bible anywhere?”
The Bible is my studio's rulebook, the file every agent reads first. My agent had the how-to written down that same night, so the next agent in that job can build me one too.
This is the same deal that runs through the whole course, turned around. When I talk, my agent does the polishing. When I decide, my agent does the legwork of making the decision easy: the item, the recommendation, the picture, one tidy question. Deciding is the human part, the one part I can't hand off. So I want it quick, and I want it clear.
If you'd have to scroll for it, it isn't in front of you.
Whenever your agent needs your OK, the exact thing you're approving goes in the same message as the question, right beside it, as the last thing you read. One question at a time, with its recommendation first.
More than a handful waiting? Ask for a form.
Make your agent bring the decision to you.
Two pastes and one habit. You'll feel the difference the first time it asks.
- Give your agent the rule.
Paste it into a new chat, or into your AI's custom instructions or project settings so it reads it every time.
Whenever you need my approval or a decision, put the exact thing I'm approving (the full text, the numbers, the list, the picture) in the same message as the question, right beside it, as the last thing I read. Never say "see above" or point me to an earlier message or a file. Ask me one question at a time, and give me your recommendation first, with one line on why. - Push back on “see above.”
The next time an agent says “see above,” “as shown earlier” or “the draft I shared,” don't go hunting. Reply: “Put it right here, with the question.”
- When questions pile up, ask for a form.
If your agent can't make a fill-in form, ask for a numbered list in the same shape.
Gather every question you're waiting on me for into one simple form I can fill in and submit, most important first. On each card: the question in plain words, your recommendation, one or two alternatives, a box for my own words, and anything I need to see or hear to decide (the text, the picture, the clip) right on the card. If you can't make a form, give me a numbered list in the same shape. - Answer in one sitting, in your own words where it counts.
Taking the recommendation is fine. That's what it's there for. But where your gut says no, write a sentence, the way you'd say it out loud. That sentence is where your judgment goes in.
Make your agent bring the decision to you: the exact thing right beside one question, its recommendation first, and a form when there are lots.
Show Me the Real Thing
Your attention is the scarcest thing you've got. No sketches, no plumbing tests: agents check their own pipes and bring you something worth judging.
Every exchange on this page really happened and is quoted exactly, slips included; […] marks where I trimmed. Times are Eastern. My agent here is Claude, made by Anthropic.