For a while, the whole game was writing the perfect prompt. Not anymore. AI tools are good enough now that the bottleneck has shifted — from getting AI to make something, to getting it to make the right something.

In this edition: why direction is quietly becoming the more valuable skill, and two habits that'll get you there faster than any prompt formula.

Remember when getting a good result out of AI felt like cracking a code? The right prompt formula, the right keywords, the right negative prompts to avoid the weird stuff.

That's not really the game anymore. AI tools have gotten good enough that a simple instruction gets you something usable. The harder part now isn't getting AI to make something — it's getting it to make the right thing.

Here's the skill that's replacing prompt-writing, and how to actually practice it.

Think about briefing a designer or a junior writer. You wouldn't hand them one instruction and expect a finished piece back. You'd give them a brief, look at the first draft, and say things like "the idea's right, but the opening is too generic" or "this doesn't sound like us."

That's direction, and it's how AI works best too. Instead of chasing the perfect single prompt, you're running a loop: brief, generate, review, give feedback, refine. The first output was never meant to be the final one — it's just something to react to.

Give it a real brief, not just an instruction.

A prompt tells AI what to make. A brief gives it the context to make better calls on its own.

"Write an Instagram post promoting my baking class" gets you something generic. "Write a post for beginners nervous about baking with yeast — warm, encouraging, not salesy, aimed at getting people to ask about the next batch" gets you something usable.

The difference isn't length. It's giving AI a goal, an audience, and real constraints instead of just a task.

Say what's actually wrong, not that it needs to be "better." This is where most AI workflows stall. "Make it more engaging" gives AI almost nothing to work with. Naming the actual problem does:

AI can hand you ten versions in the time it used to take to make one. Choosing the right one is still on you.

A technically solid piece of content can still be:

AI generates the options. You decide which one is worth using — and as generation gets faster, that call matters more, not less.

Say you're making a Reel. Version one has a good idea but a flat opening. You don't start over — you tell AI to keep the concept but rework the opening so it creates curiosity in the first sentence. Version two is closer, but the ending feels salesy, so you ask it to keep the CTA but make it feel like a natural next step instead of a pitch.

That back-and-forth is the whole skill. You're not trying to predict every detail upfront — you're shaping the work as you see it.

Next time, skip the 20 minutes spent perfecting one prompt. Give AI a solid brief, let it produce a first version, then ask yourself:

Feed that back in. Repeat.

The tools will keep changing. Knowing what you want and how to get closer to it won't.