If you still write prompts that tell AI exactly how to think, you may be making it less intelligent. The smarter models get, the more that old habit gets in the way. Here’s the switch.
The prompting advice everyone handed you two years ago is going stale. Follow it today and you might be making your AI worse.
Here’s the guide in one line. Stop telling AI exactly how to think, and start telling it what you want. Writing out every step used to help a lot. With today’s models, it mostly gets in the way.
A couple of years ago, models needed hand-holding. You’d spell out every move, from analyzing the text to summarizing it to rewriting it in your tone. You were the brain and the AI was the hands. If you left a step out, it would wander off.
Today’s models, like Claude Opus 5.5 and Fable 5.1 or OpenAI’s GPT-6, plan, check their own work and find a smart path on their own. Anthropic’s own prompting guide now says to "prefer general instructions over prescriptive steps," because "Claude’s reasoning frequently exceeds what a human would prescribe." Spell out every tiny step and you box in the reasoning you’re paying for.
How vs What
A how prompt lists the pieces in order, like a headline, then three bullets, then a short paragraph and a button label for my landing page. A what prompt says "I need a landing page that gets busy founders to book a demo. Here’s my product and my audience. Build me the page." The second gives the model room to do its best thinking.
The simple way to remember it: prompt engineering describes the process. Intent engineering describes the outcome. The old way says here’s exactly how to solve this. The new way says here’s the result I want, you figure out the best route.
You go from the micromanager dictating every keystroke to the person who sets a clear goal and trusts a capable team to hit it. On a weak model, dictating steps was a safety net. On a strong model, it caps how good the answer can be.
Each rewrite drops the steps and names the result instead.
01
Competitor pricing
Before: "Research my competitors and list their prices, then compare them in a table."
After: "I want a clear picture of how my pricing stacks up against my top competitors. Show me what matters most."
02
Blog post
Before: "Write an intro and five tips, then a conclusion."
After: "Write a blog post that makes a nervous beginner feel confident about starting. You decide the shape."
03
Meeting notes
Before: "Summarize this and pull the action items, then format them as a list."
After: "Turn this meeting transcript into something my team can act on tomorrow morning."
04
Tone
Before: "Use a formal tone with short sentences and no big words."
After: "Write this so a smart twelve-year-old gets it on the first read."
Outcome-based does not mean vague. "Make me a website" isn’t intent engineering. It’s a lazy prompt, and you’ll get a lazy result.
You still give rich context and real constraints, like a word count, a budget, your brand voice or a rule it can’t break. You still show an example of what good looks like, and you still say what done means. And if a step truly has to happen in order, like a disclaimer that must come first, keep it as a rule. The only thing you drop is the step-by-step method.
Think of it like briefing a brilliant freelancer. You give them the goal, the context and the guardrails, then you get out of the way.
Here’s the converter from the video. Paste in any of your old how prompts and it hands back an outcome-based version.
Your Homework
Run your five favorite prompts through the converter this week. Every time you catch yourself telling AI exactly how to solve something, ask if you could describe the result you want instead.