The Prompt Is Not The Point

Most men my age are waiting to be retrained or reskilled. I personally think that they will be waiting a very long time.

Nobody is coming. There is no course, no corporate rollout, no two-day workshop that will do this for you. I have spent nearly thirty years in civil engineering and transport planning, on several continents, and I can tell you that the single biggest determinant of whether someone my age gets any value out of AI is not intelligence, technical background or budget. It is the willingness to look stupid for about six weeks.

I looked stupid for about six weeks.

Stage one: writing wishes

My first prompts were wishes. “Summarise this report.” “Write me a market entry strategy.” “What are the benchmarks for metro construction costs?”

I got mush. Confident, well-structured, plausible mush — the sort of thing a very keen graduate produces on their first Friday when they have not asked a single question all week. And like most people, I drew exactly the wrong conclusion. I decided the tool was overrated.

It wasn’t overrated. I was under-briefing it.

Think about how you actually delegate. You do not walk up to a new starter and say “do the rates benchmarking.” You give them the contract. You tell them the client, the currency, the seniority tiers, the format the PMO expects, the three things the last submission got wrong. You tell them what good looks like. Then you leave them alone.

I was doing none of that. I was typing one sentence into a box and expecting a deliverable.

Stage two: context is the whole job

The turn came when I stopped writing prompts and started writing context.

Markdown files. Plain text, structured with headings, bullets, tables. Nothing clever. One file describing the project — the site parameters, the institutional landscape, the standards that apply, the terminology. One file describing how I want work produced — Word documents with source citations and confidence ratings, no unsourced assertions dressed as fact, professional judgement flagged as professional judgement. One file holding what we have already established, so I never have to explain it twice.

That is it. That is the whole trick. Markdown files that travel with the work.

The difference is not incremental. It is the difference between a stranger and a colleague of three years. When I open a benchmarking task now, the model already knows that unit cost rates are in SAR, that the deliverable is a colour-coded workbook with one firm per tab, that the naming convention is fixed, that a triangulated estimate must carry an explicit confidence caveat. I do not restate it. It is written down.

Write the context down. Refuse to repeat yourself. Every sentence you type twice is a file you should have written once. If you’re already there, maybe start thinking about creating skills so that you can streamline your processes in the future.

What it actually does

Three examples from the last year, all real, none of them toys.

Benchmarking. Infrastructure CAPEX work using actual outturn costs as the primary comparator, cross-referenced against the NYU Marron Transit Costs Project and the Flyvbjerg cost-overrun literature. Actual detailed comparisons of publicly available, exemplary mobility and transport hubs and pulling out details that are pertinent to my work, and that a human would normally miss from the sheer amount of data. Work that used to be a fortnight of a junior’s life, done properly in an afternoon, with the sourcing hierarchy visible so I can audit every number.

Go-to-market. Pitch narrative and market sizing for NAYRobotics, my warehouse automation venture. A Gulf-focused market entry due diligence pack for a Madrid consultancy — market conditions, competitive landscape, MISA licensing, Saudization, financial estimates, twenty-three slides and a full Word appendix. Drafted with me, not for me.

The virtual PA. Inbox and diary. Email triaged and drafted, meetings scheduled, documents filed to Drive, the CRM kept current. Every outbound draft is flagged DO NOT SEND until I have read it myself. That rule is not negotiable and never will be.

Where it will burn you

It lies. Beautifully.

It will invent a citation with a plausible author, a plausible journal and a plausible year, and it will do it in the same even tone it uses for the things that are true. It will produce a number that is directionally sensible and factually fabricated. If you cannot tell the difference, it will eat you.

This is precisely where thirty years is worth something. I know what a plausible day rate looks like for a Grade 3 engineer in the Gulf. I know when a cost-per-kilometre figure has drifted into fantasy. I know which standards actually apply and which ones the model has hallucinated into relevance. The judgement is the scarce input. It always was.

So verify. Demand sources. Make it separate verified citation from professional judgement, explicitly, in every document. Then check the citations anyway. Create feedback loops like I’ve done, and if you’re really getting competent with your prompt engineering, you can even create independent AI checkers with explicit checking criteria (like a really good QA/QC person).

The part nobody says out loud

The advantage in this is not with the twenty-five-year-olds. It is with us.

The judgement is the scarce input. It always was.

They can prompt faster. Fine. But prompting is not the point — the prompt is the least valuable part of the entire exercise. The valuable part is knowing what question to ask, what a good answer looks like, and when you are being told something that is confidently wrong. That takes decades. You already have the decades.

What you do not have is an excuse.

Start with one real piece of work. Not a toy, not a demo — something you actually have to deliver this month. Write down the context first. Give it the brief you would give a competent new starter, then set the guardrails to ensure that your data isn’t misused or co-opted to train the AI provider’s models. Read the output like a reviewer, not an audience.

Then do it again tomorrow.

Show up, do the work, go home. Same as it ever was. The tools changed. The hard graft didn’t.