ONE-LINE STING: Being clearer about what you want has always been a skill. Paying £120,000 for it is new.
"I've been writing prompts for forty years. We called it talking."— Office Man
Prompt Engineering: Is It Actually a Job?
May 2026 · Issue Two
WHAT IS IT?
Prompt engineering is writing better instructions for AI — with enough specificity, context, and structure that the output is genuinely useful rather than vaguely adjacent to what you wanted. At its simplest, it's the difference between "write something about pensions" and "write a 200-word plain-English explanation of auto-enrolment for new employees who've never thought about retirement savings." The second brief produces something you can actually use.
Techniques range from basic (be specific, give context) to intermediate (give an example of good output, ask the AI to reason step by step) to specialist (structured system prompts, automated testing across variations). Most people need the first two levels. The specialist tier is for teams building AI products.
WHY NOW?
In 2023, early AI models were sensitive to exact phrasing in ways that made skilled prompting feel like an almost technical skill. Job listings appeared. Courses launched. The gap between a vague request and a well-structured one produced measurably different results.
Since then, models have improved substantially at understanding intent. Some of the workarounds that mattered then matter less now. But the underlying principle hasn't changed: AI tools work better when you brief them properly. The hype around the job title has faded. The practical skill remains.
THE HUMAN ANGLE
Most people who feel AI "doesn't work" are using it the way they'd use a search engine — a few keywords, take what comes, repeat. Most people who find AI genuinely useful have quietly developed a briefing habit: context first, format second, constraints third.
This isn't a generation gap or a technology gap. It's a communication habit. The people who are good at briefing colleagues, writing clear emails, and specifying what they actually need tend to get better AI outputs with almost no training. The skill transfers. It just needs applying.
Worth trying once: Add "think through this step by step before answering" to any prompt involving analysis. It makes a real difference on complex questions.
WHAT TO DO
Before you type, take ten seconds: what exactly do I want, who's it for, what format, what constraints? Paste in an example when you want a specific style. Treat the first output as a draft, not a final answer. Iterate with specific feedback rather than starting over. And skip the courses — Anthropic's free documentation covers everything a non-specialist needs in under an hour.
The goal isn't mastery. It's the habit of being clearer about what you want. Which, as it turns out, is useful everywhere — not just with AI.