Process before prompt
Don't focus on the end, focus on the journey first
You decide to bake a cake (don’t ask me why…you could just grab one at the store). You download a recipe and dive in. Two minutes later, you’re stuck. You figured: I want cake > I need a cake recipe. Buuuut that’s not how it works.
Before you even switch on the oven, you’ve got to know: what kind of cake am I making? What goes in it? How many people is it for? How does my oven work? How much time do I have? Without that info, even the fanciest recipe won’t save you.
Same with AI. People think: I want to write an article > I need a prompt that says “write article.” Just like with that cake, you’re missing half the ingredients and all the context. That’s how you end up with prompts that are either crammed full of random instructions or so bare they collapse. Also: the output sucks.
Instead of obsessing over the final product (still important!), good prompting means first looking at your process. Surfacing all the little things you normally do on autopilot and naming them.
Why bother?
Whether you’re prompting by hand, building agents, or cooking up something else “magical”: mapping out your task process helps you see, step by step, if AI makes sense here and whether it’s even allowed (high-risk steps may be off-limits). It also shows you what you actually need to move each step forward: approval, feedback, a briefing, whatever.
So your prompt adventure doesn’t start with the prompt. It starts with your process. And yeah, you might think: bro, I know exactly what steps I take to write that article. I guarantee there are more than you realize. You’re good at your job, so it feels like one smooth move. But zoom in and it’s twelve micro-steps and hidden detours. You’ve got to tease those apart before you can throw a prompt at them with any sense.
How do you do that?
When I train people in prompt design, I never start with the prompt. That’s like preheating the oven and then shoving in half the ingredients, in the wrong amounts, in a plastic tub. That’s not a cake, it’s a crime scene.
So first we go through some questions. I knowww… snooze. We’re doing it anyway. In a workshop I’ve got hours, exercises, and sassy slides, but here you’re only giving me a minute, so let’s cut to the essentials:
What steps do you actually go through? Spell them out.
What do you need to start and finish each step? Think briefings, sources, data, examples, format, channel, audience, access, owners.
Can and may AI help here? Is this a check, a draft, a rewrite, a summary, a classification, a decision note? “No” is a valid answer. And if your answer is “maybe,” that’s a research hint.
LLMs can’t do everything, and definitely not everything equally well. Be picky about what you hand over. Not just to write sharper prompts, but to underline where you are still essential in the process. In a time when everyone keeps saying our jobs are on the line, that’s not a bad side effect.
What’s in it for you?
You start to see where the real work lives: in the micro-steps. You stop AI from guessing, which keeps hallucinations at bay. You spot where your judgment is critical (stays with you) and where repetition can speed up (hand it off to the machine). And you get a clearer view of which inputs, formats, owners, and decision points need to be set up upfront, so fewer endless feedback loops.
First exercise for tomorrow
Pick one task you do all the time. Write down the steps: talk to a subject-matter expert, clean up notes, check if you have enough input, whatever. Answer the three questions for each step. Then grab the ones GenAI can clearly handle and write prompts for those (or agents….whatever floats your boat). Your prompts will come out sharper, cleaner, and way more effective.
Say it with me: process before prompt.
Prompting isn’t magic, it’s mise en place. Without a recipe, you’ll keep serving raw dough. With a recipe, you’ve got insight and control. Prompting doesn’t start with the end product; it starts with knowing your process.
