Brief AI like a senior consultant

A team lead needed to hire a backend engineer. He opened an AI chat and typed: "Write a job description for a backend engineer."

In five seconds he had one. "Fast-paced environment." "Rockstar problem-solver." "Competitive salary." A list of ten technologies, half of which the team didn't use. It looked like every other job post on the internet.

He posted it anyway. Two weeks later he had sixty applications and not one he wanted to call.

So he tried again, this time with a real brief. We're a team of five. We work in Ruby on Rails. This person will own our payment flow, which breaks at the worst moments. Success in the first 90 days means our checkout errors drop and nobody gets woken up at night. We work remotely and write a lot. Good candidates hate buzzwords, so don't use any.

The new post was half as long. It got a third of the applications, and most of them were worth a call.

Same AI. Same job. The only difference was the brief.

Brief it like you're paying by the hour

Most weak AI output doesn't come from weak AI. It comes from weak instructions.

Here's the rule of thumb I use: brief AI as if you were paying a senior consultant by the hour. You'd never hand them one line and walk away. You'd tell them who the audience is, what they care about, what has worked before, and what success looks like.

Before you hit enter, check four things:

• Who is the reader, and what pressure are they under?
• What tone will they respect?
• What proof or references can you give it to work with?
• What exact outcome do you want?

People call this context engineering. I call it briefing properly. It often makes the answer several times better, with the same tool.

Make it show its thinking

Context fixes the input. The second habit fixes the reasoning.

When you ask a hard question, AI tends to jump straight to a neat answer. Ask "How do I handle a conflict between two people on my team?" and you'll get a sensible-sounding framework. But you can't see what it assumed, so you can only trust it blindly or throw it away.

Instead, ask it to think out loud first:

"Before you suggest anything, walk me through your thinking. What might the root causes be? Who has more formal and informal power here? What happens if I do nothing? What does a great outcome look like? Then give me a plan."

Now you can see where its picture matches reality and where it's wrong. Correct the wrong step, and the plan gets better. You're not just reading an answer anymore. You're shaping the thinking together.

This is often called chain-of-thought prompting, and it works anywhere the reasoning matters more than a quick headline: hiring, budgets, negotiations, hard conversations.

The tool isn't the bottleneck

Everyone has access to the same AI now. What separates good results from average ones is the person writing the brief.

Give it the context you'd give a smart colleague. Ask it to show its work. Push back.

That's how you move from Co-Pilot to Co-Thinker.