"Does it know what I actually need?"
Description is the craft of communicating with AI. It is also where most professionals' frustration with AI actually lives. Most people who tell you AI "isn't very good" have a Description problem, not an AI problem. They have asked for something — but they haven't conveyed what they need. The context, the audience, the constraints, the tone, the stakes. AI infers what it can. Without those signals, what it infers is generic.
The way you brief a colleague is the way you should brief AI. A good brief includes who this is for, what success looks like, what format is needed, what to avoid, and any background the other party couldn't reasonably be expected to know. None of this is unusual — these are the exact questions a good agency creative would ask before they touched a piece of work. The shift is recognising that AI deserves the same input. Garbage brief, garbage output. Precise brief, precise output.
The single biggest unlock for most professionals is learning to give AI not just a task but a context. That context is your expertise, your situation, your standards, your house style. Once that context is established — once AI knows who you are, what you do, who you're writing for, and what good looks like — every subsequent prompt sits inside that frame. The work compounds. Each prompt produces better output because the surrounding context is doing half the work.
Description is where prompt engineering as a craft sits. It matters. But it sits inside the larger discipline of AI fluency, not above it. Knowing how to write a precise prompt is useful only if you've already made the right Delegation call. And it produces little value if Discernment and Diligence aren't there to evaluate and own what comes back. Description is one quarter of the picture, well executed.
For senior professionals, the practical move is to invest once in setting up persistent context — a project brief, a system prompt, a set of style examples — and then let every subsequent interaction inherit it. AI brings the speed. You bring the meaning. The brief is how the meaning gets transferred.
Asking AI for things the way you'd ask a search engine. No context, no audience, no constraints. The output is technically responsive and practically useless. The fix is structural, not technical: brief like you would a capable junior colleague, every time.
Specifying so tightly that AI has no room to surface what you didn't already think of. Description should narrow the gap, not eliminate the conversation. Leave AI room to push back, ask clarifying questions, or offer alternatives.