Prompting and Practical Workflows

Ask ten different people to use an AI tool for the same task and you'll get ten very different experiences — not because the tool is inconsistent, but because of what they asked for and how they asked for it. Prompting is a learnable skill, and the gap between someone who prompts well and someone who doesn't is significant in terms of the quality of output they receive.

This matters more in a professional context than a casual one. A vague prompt in a personal setting means a slightly off answer you can ignore. A vague prompt in a professional context means a draft you have to almost entirely rewrite, or an analysis that misses the point of what you actually needed.

There's a common misconception that good prompting means finding "magic words" or specific tricks. It doesn't. Good prompting means communicating clearly — the same skills you use when briefing a colleague, writing a project scope, or explaining a problem to a new team member. The difference is that with AI, you have to provide context that a colleague would already know from working with you.

Think of it this way: if you hired a highly capable contractor with no context about your project and asked them "can you help with the thing we talked about?", you'd get a confused response. Give that same contractor a clear brief — who you are, what you're working on, what specifically you need, what constraints apply — and they'll produce something useful. AI is the same.

A well-structured prompt has four components. You don't always need all four — simple requests can be simple — but understanding each one helps you diagnose why a prompt isn't working and fix it.

You won't always need all four sections. A simple request like "explain what a webhook is in plain English" needs no role, context, or format instruction. But for any professional task — a draft document, an analysis, a set of options — building in these components will dramatically improve what you get back.

Beyond the basic four-component structure, there are several prompting patterns that are particularly useful in professional IT contexts. These aren't tricks — they're ways of giving the AI better information so it can produce better output.

Pattern 2: Iteration. Treat AI conversations as a dialogue, not a one-shot query. If the first response isn't quite right, don't start over — refine. "That's good but too formal for this audience — can you make it more direct?" or "The summary is too long — cut it to the five most important points" or "Add a section on risks we haven't mentioned yet." Each iteration costs you seconds and improves the output significantly.

Pattern 3: Ask for options, not answers. Rather than "what should I do about this architecture decision?", try "give me three different approaches to solving this integration problem, with the tradeoffs of each." You get richer material to work with and your judgment decides which direction to take. This is particularly valuable for decisions where context you hold (political, relational, historical) is as important as technical considerations.

Pattern 4: Provide examples. If you need output in a specific style or format, show the AI what you mean rather than trying to describe it. "Here's an example of a requirements statement we use on this project: [example]. Write five more in the same format covering [topics]." Matching existing formats and styles is something AI tools do very well when given a reference.

The most effective AI users aren't the ones who use it for everything — they're the ones who've identified the specific points in their workflow where it creates the most value and built it in there deliberately. The goal isn't to use AI more. It's to work better.

Here are the highest-value integration points for IT professionals, organised by the type of work:

What to avoid: Using AI for tasks where accuracy is critical and verification is difficult. Specific client data, regulatory requirements, legal commitments, or technical specifications that will be used in production systems — these need primary sources, not AI synthesis. This isn't a reason to not use AI; it's a reason to be clear about when you're in draft-and-review mode versus reference mode.