Your AI can do almost anything. The hard part isn't capability - it's understanding. Intent engineering is the discipline of translating human intention into machine-executable clarity.
Prompt engineering tells the AI what to do. Intent engineering tells it why.
The Intent Gap
When you ask for "better," you have a rich internal model of what better means in this context. The voice you want. The audience you're targeting. The constraints you're operating under. None of this is in the prompt.
The model receives your words and fills missing context with defaults - statistical patterns from training. Its defaults aren't your intent. It doesn't know what you meant, only what you said.
The output is technically responsive to your prompt. But it's not what you wanted. The gap between intent and instruction produced the gap between expectation and result.
What Intent Engineering Means
Intent engineering is the practice of making explicit what you usually leave implicit. The audience. The tone. The constraints. The success criteria. The anti-goals - what you explicitly don't want.
Prompt engineering optimizes words. Intent engineering optimizes the entire context transmission. Examples. Counterexamples. Rubrics. Reference points. The prompt is just one channel.
Intent rarely transmits perfectly on the first try. Intent engineering includes the iteration protocol. How do you correct misunderstanding? How do you refine without overwriting?
Intent Engineering In Practice
Before any task, answer: "Why does this matter?" Not what you want done - why you want it done. The purpose shapes every downstream decision.
Who will consume this output? Their expertise level, their context, their needs. "Write clearly" means different things for executives vs. engineers vs. customers.
What do you NOT want? The AI will optimize toward what you measure. If you don't specify what to avoid, it will include things you hate. Anti-goals are as important as goals.
Examples of good output. Examples of bad output. Reference points that show, not tell. "Like this but shorter" transmits more intent than paragraphs of description.
How will you know if it worked? Not vague ("good") but specific ("the reader can act on this without asking follow-up questions"). Testable criteria close the intent gap.
Where Intent Gets Lost
You know so much about your context that you forget to transmit it. What's obvious to you is invisible to the model. Assume nothing is shared.
You ask for one thing but measure another. "Make it engaging" but you evaluate by word count. The model optimizes for what you actually measure. Align your metrics with your intent.
"Be creative." "Think outside the box." "Make it pop." These feel like instructions but they transmit zero actionable intent. Abstract requests produce random results.
Each revision adds constraints without removing old ones. The intent becomes contradictory. The model contorts itself to satisfy incompatible requirements. Sometimes you need to restart clean.
Developing Intent Fluency
Before engaging AI, write down what you actually want. Not the task - the underlying need. If you can't articulate it to yourself, you can't transmit it to a model.
When you revise AI output, notice what you're changing. That's the intent gap made visible. Next time, transmit that intent upfront.
Collect successful intent transmissions. The prompts that worked, the context that clarified, the examples that anchored. Intent engineering is a craft with reusable patterns.
The AI revolution gave us capability. Intent engineering gives us control. The models can do almost anything - the constraint is making them understand what we actually want.
This is a new discipline. Not prompt engineering - that's syntax. Not AI strategy - that's organizational. Intent engineering is the cognitive work of translating human purpose into machine understanding. It's the missing layer.
The prompt is what you said. The intent is what you meant. Close the gap.