Before I let AI write this newsletter, I had it interview me.
One question at a time. I answered by talking, the same way I do when we start a long-term project together.
During that conversation, we got to a step I hadn't included in my original request: the first interview should build the project's standing instructions. How we work together comes before what we're going to make.
My first defense against AI slop is an interview.
If you're getting generic copy and cycling through corrections, look at what the AI had to work with before you blame the tool.
You know the customer and the sales pitch you refuse to use. You know what failed last year. You can spot an example that sounds nothing like you.
The AI needs that information too.
You pay for missing context in revisions.
OpenAI's own prompting guide recommends including relevant context. The interview is how I get mine out of my head.
It won't prevent every bad answer. You still have to check the work. But correcting a decision the AI never knew you'd already made is avoidable.
I'd rather spend tokens explaining the work before it starts.
Build the instructions before the deliverable
At the beginning of a long-term project, have the AI interview you about how you want to work together.
Define what good work looks like, how direct the feedback should be, and which decisions need your approval. Set rules about facts it must verify and things it must never invent.
Those answers become the project's standing instructions, the rules you want it to follow throughout the work.
I tell my agents to take a Gen X, no-nonsense, blue-collar Wu-Tang mentality. Don't mince words. If an idea doesn't make sense, call me out and tell me why.
Automatic agreement doesn't help me refine an idea.
Pick your own working style. But give the AI permission to question a vague goal or a plan that doesn't solve the problem you described.
Then interview for the project itself. Get clear on the audience, outcome, and constraints. Define how you'll judge the result.
For a sales project, go beyond asking for a sequence of emails. Explain who buys and what makes them hesitate. Supply evidence for what you can honestly promise.
Otherwise, you're asking the AI to fill in details that belong to your business.
Answer out loud
This works better for me when I talk through the answers.
I give the AI far more information by speaking than I would type. A follow-up question gets me thinking about something I hadn't considered. Talking gives me room to work through it.
I recommend Wispr Flow for this. It turns what you say into text you can put into the conversation.
Use dictation you already have if that's easier. You shouldn't need another purchase to explain your own project.
Ask for one question at a time. Answer it. Let the AI follow up where something is vague.
Then make it show you what it understood.
Before any execution, review a concise project brief and the proposed instructions. Correct the gaps and approve the direction.
If the summary is wrong, the AI isn't ready to work.
Steal this kickoff prompt
Paste this into your AI tool before your next long-term project. Replace the bracketed text.
I'm starting a long-term project: [describe the project and what you already know].
Before doing the work, interview me to build the project instructions and brief. Ask one question at a time, wait for my answer, and adapt your follow-ups. Don't give me a questionnaire all at once or repeat what I've already answered.
First, learn how I want us to work together: my standards, preferred communication style, how directly to challenge me, boundaries, and decisions that require approval. Ask for examples of work I like or dislike when useful.
Then establish the project's goal, audience, current situation, desired deliverables, resources, constraints, deadline, and success criteria. Ask for the facts and source material you need instead of inventing them.
Challenge vague answers and contradictions. If you disagree, explain why. Separate confirmed facts from assumptions. Don't agree merely to be agreeable or argue for the sake of it. Ask only for information relevant to the work; flag sensitive material before requesting it.
When you can summarize the goal, audience, scope, constraints, and success criteria, give me the project brief and draft reusable instructions I can save. Flag unresolved questions and assumptions, especially anything that could change the plan. Stop and wait for my corrections or approval before producing deliverables or taking action.
Start with the most useful first question.
Save the approved instructions in your tool's project instructions area or as a reference file. If it doesn't have a project feature, keep the brief and paste it into future sessions. Don't assume a new chat knows what you settled elsewhere.
Try this on your next project. Read the brief before you let the AI start.
My boy Chris keeps showing up
Chris Paradiso and I met around 2007 or 2008.
He's been there ever since, with encouragement and tough love when I needed it. The man calls, texts, and sends memes to keep me moving. We've done that for each other for a long time.
You don't find many people who keep showing up, year after year.
Chris is one of those people for me.
That's why I'm so proud of him for publishing Civilized Savage: The Daily War Against Comfort. I know how honest he is. I know the lessons in that book come from the way he lives.
He wrote it for his son and daughter, to help them understand the experiences and decisions that made him who he is. As a dad, that matters to me.
Our new Finding Peak conversation gets into doing hard things on purpose, keeping promises to yourself, and knowing who you're doing the work for.
Getting to sit down with someone who has pushed me to grow for this long, and talk about the lessons he wants to pass on to his kids, meant a lot to me.
Chris, I'm proud of you, brother. Thank you for continuing to show up.
From the sponsor: PFS Global
PFS Global handles the accounting, tax, and payroll work that keeps finding its way back onto an owner’s desk.
Property and casualty insurance agencies are a specialty. Every business is welcome.
I trusted PFS with accounting and tax work at Rogue Risk.
I work with the PFS team now.
If the back office is eating the time you need for clients, people, and growth, start at pfs.ryanhanley.com.
Find the work worth protecting
I'm writing Easy Mode, planned for September 2027, to help you build your life around the work you're built for. The free Edge Finder gives you a place to start today.
Start with four specific activities. Look at each through energy, contribution, demonstrated ability, and flow.
You get a map of your work, your strongest current edge candidate, and a clearer view of what deserves more attention or a harder conversation. That gives you something to test against your calendar.
Your core result is available without giving us your email address.
Before you hand AI more work, get clear on the work you want to keep.
This is the way.
Hanley.
