Structure beats cleverness
Why prompt engineering matters
A Harvard/BCG study split consultants into three groups: no AI, AI without training, and AI with prompt training. The trained group performed best. They also discovered a 'jagged frontier' — tasks where AI helps enormously, and tasks where relying on it makes performance worse.
Centaurs and cyborgs
The study found two working styles. Centaurs divide and delegate — they hand a whole task to the AI and come back when it is done. Cyborgs work in rapid back-and-forth with the model. Both styles have strengths; the key is knowing which to use when.
The four prompt ingredients
A strong prompt includes: a role (act as a renewable energy expert), a task (summarise this paper), context (it is 10 pages, for policy makers), and a format (five bullet points, key findings only). Adding these turns vague prompts into precise instructions.
Reflection improves output
Asking the model to critique its own output and then revise it is one of the most effective prompt techniques. It costs one extra step but often produces a noticeably better result.
Take any recent prompt you have used. What role, context, or format instruction was missing?