Scope, Sequence, Signal. The System That Makes AI Actually Help
Structured AI takes work off your plate. Left unstructured, it piles more on.
AI does the most damage when it answers a question you didn’t ask.
Researchers at UC Santa Barbara released a study this past March that tracked 34 finance professionals as they built a company valuation using ChatGPT. They wanted to know when AI help was actually helping, and when it was making things worse. When finance professionals incorporated the AI’s suggestions into their final reports, the reports improved. That part was expected. What the researchers didn’t expect was what happened when the AI added things nobody asked for.
Say someone asked the AI to project next year’s cash flow. The AI answers, and then keeps going. It suggests a list of similar companies to compare the business against. Useful-sounding, but not what got asked. Now the person has to read it, judge whether it’s right, decide whether to use it, and remember where they were before the detour showed up.
Every time that happened, the final report got worse. 3x worse than if the task itself had been harder to begin with. A hard task on its own didn’t damage the work. AI drifting off the question did.
AI drift is what pulls a task off course, adds work you didn’t plan for, and leaves you deciding which parts of the answer to keep. Three boundaries stop AI drift before it starts: Scope, Sequence, and Signal.
Hey, I’m Hodman. I write Between Thinking and Doing for high performers facing execution friction who want to gain traction on their complex lives.
If you're looking for a workspace that organizes itself when your executive function is depleted, I published a piece on Sunday that walks you through How to Build an AI Second Brain Using Claude + Notion.
Scope
Scope means writing down the one task you’re working on before you open the tool.
For an ADHD brain, the hard part is deciding what the task even is before you start typing. You know the general area you’re working on. You haven’t picked which specific part of it you’re working on right now. You start the conversation anyway, and the AI reads your general prompt and picks a specific piece for you.
Now you’re solving two things at once: the task, and whatever the AI thought the task was.
Writing the task down before you open the tool stops that from happening. When the AI gives you an answer, you can read your written task, read the answer, and see whether they match. Without that written task, you have nothing steady to compare the answer to.
Sequence
Sequence means asking for the short answer first, and only asking for more if you need it.
AI defaults to giving you the long version. Ask for a plan, and you get the plan, the reasoning behind the plan, three alternatives, and a list of things to watch out for. All of it looks helpful. All of it now has to be read, judged, and either used or ignored before you can move.
Asking for the short version first changes what you get back. You get the one answer you asked for. If you want more, you ask for more. Try this prompt:
Give me the shortest useful answer first. Only give me more detail if I ask for it. Follow this pattern for the rest of our conversation.
Paste that at the start of the chat, and the tool will follow that pattern for the rest of the conversation.
Signal
Signal means catching the moments when the AI adds something you didn’t ask for, and deciding right then whether to keep it or throw it out.
In a single conversation, you can usually catch drift as it happens. You notice the AI added something you didn’t ask about, and you pull it back. Across a project with multiple threads open over multiple days, catching drift by memory stops working. You lose track of which conversation said what, and what you decided to do about it at the time.
That’s a tracking problem. Tracking problems need a system.
Coming Up
Scope, Sequence, and Signal work together. Two pieces this week give Sequence and Signal the room they need to help you.
Sequence is really about trusting a short answer over a complete one, which is the subject of this week’s follow-up, ‘Stop Trying to Make It Impressive. Good Enough Gets It Done.’
And when Signal has to work across a project with multiple threads instead of one conversation, you need the workflow for it. That’s what ‘Stop Losing Track of Who Said What. Use AI to Manage Complex Projects Instead’ walks you through, published at the end of this week.