Read More
We have calibrated our understanding of AI's nature, potential, and limits. Now the question shifts from what AI is to what humans must do before using it.
ADVERTISEMENT
SCROLL TO CONTINUE WITH CONTENT
Most AI failures do not begin with the model. Nor do they begin with the prompt. They begin before the prompt, when the human has not yet decided what success means. Generation is now frictionless, but specification is not. The result is a new pathology of the AI age: premature generation. We start producing before we have decided what deserves to be produced. The machine fills the vacuum with plausible structure, and the work drifts toward a problem nobody consciously chose. It is like shooting arrows before choosing the target. The activity is real, but the aim is missing.
Specification is the discipline that resists this drift. Before any prompt, it requires the human to decide five things: the objective, the scope, the constraints, the context, and what "done" looks like. None of these are prompt mechanics. They are acts of thought that AI cannot perform on behalf of someone who has not yet performed them.
The most useful habit in the AI age may therefore be the opposite of what most people practice. Before asking AI to generate the answer, ask AI to generate the specification. Have it propose the objective, name the constraints, surface what is missing. The first output is not the work. It is the brief for the work, written collaboratively, so that when generation begins it is already aimed.
This also trains the human. Each specification cycle exposes vague intention, hidden assumptions, and missing context. Over time, the human becomes better not only at prompting AI, but at defining problems.
AI does not remove the thinking. It moves the thinking upstream, before the prompt. The reinvented human does not begin by typing better. They begin by wanting more clearly.
Frank Ng is a retired NASDAQ CEO, who co-authors this column with his son Ryan after publishing their book Hey AI, Let’s Talk!
