Most prompting guides start from the same assumption: you know what you want, you just need to say it better. Add more detail. Be specific. Use a framework.
That assumption is wrong.
The real problem is upstream. Most experienced professionals don't consciously know what makes their work good. They've been doing it so long that the judgment calls, the quality filters, the trade-offs they navigate automatically have all gone underground. They can't articulate those things to a colleague, let alone to a machine.
Telling someone to "just tell AI what you want" is like telling them to explain how they ride a bike. The knowledge is real. It is just not in words yet.
Three tiers most people don't see
The conversation about AI communication is stuck on the first tier. Prompting is about what to say: word choice, structure, role assignment. It is where every tutorial starts and where most people stop.
The second tier is context engineering: what information to provide. Background, constraints, audience, examples. This is where real improvement starts, and some organizations are getting here.
The third tier is intent engineering: what the AI needs to want. Not the task. The purpose behind the task. The definition of "good" that lives in your head but never made it into the instruction.
Have you ever looked at AI output that was technically correct and felt something was off, but couldn't name what? That gap between "correct" and "right" is intent. It is the tier almost nobody is working on.
What happens when intent stays invisible
Klarna deployed AI across their customer service operation with a clear, measurable objective: keep call times low. The AI executed brilliantly. Resolution times dropped. Metrics improved.
And the CEO publicly admitted quality suffered. They started rehiring humans.
The AI did exactly what it was told. The problem was that the actual intent (build lasting customer relationships, make people feel heard, solve the real problem) was never stated. It lived in the heads of experienced service reps who knew that sometimes the right answer takes longer. That knowledge was invisible to the system that replaced them.
This is not an AI failure story. It is a delegation failure story. The same thing happens when you hand a project to a new hire without explaining what "done well" looks like.
A skill older than computers
Intent engineering sounds technical. It is not. It is delegation. Good managers have been doing it for centuries: making their standards explicit, defining what success looks like, naming the trade-offs that matter.
Stop and think about the best boss you ever had. Chances are, they were good at telling you not just what to do, but why it mattered and what "good" looked like. They externalized their intent. AI demands that skill of everyone now, not just managers.
The bottom line
The prompting conversation is a distraction from the real work: surfacing the judgment, standards, and trade-offs that experienced professionals carry without thinking about them. The next time AI gives you something wrong, don't rewrite the prompt. Ask yourself: "What did I know that I didn't say?" That question is where intent engineering starts.