The Part of AI Development Nobody Talks About

The Part of AI Development Nobody Talks About

Everyone talks about which AI tool to use.

Nobody talks about how to tell it what to build.

I spent six months trying different tools before I realized that was the wrong problem. Claude, GPT-4, Cursor, Copilot — they are all good. They are all capable of building what I needed.

The bottleneck was me. Specifically, my inability to describe what I wanted with enough precision for the AI to act on it.

Here is what I mean.

When I say "I want to add a login to my app," I think I have been clear. I have described a feature. What I have not described is:

• What kind of login? Email/password? OAuth? Magic link?
• What should happen after login? Dashboard? Onboarding flow?
• Do I have existing auth infrastructure or are we starting from scratch?
• What happens if the login fails?
• What page does the user come from? What page do they go to?

Every one of these blanks will be filled by the AI, using defaults I did not choose.

A developer who builds login systems all day has good defaults. They are not the same as your defaults.

The way I fixed this was a process I now call brief-first. Before I type any request, I write out answers to four things:

What I am building — one specific thing, described specifically enough that I could hand it to a contractor.

What I already have — stack, existing files, existing integrations. Anything the AI should build on instead of around.

What it should NOT do — explicit constraints. This is the most important part. "Do not change files outside what I mentioned. Do not install packages. Do not add features I did not ask for."

What success looks like — the exact test I will run to know it worked.

The first time I tried this, I sat down to write a brief for a feature that should have taken thirty minutes. The brief took fifteen minutes to write. I realized halfway through that I had not thought through two important edge cases.

Those edge cases would have cost me three hours of revision cycles if I had discovered them in the AI output instead of in my own thinking.

The brief is not overhead. It is the work. The AI does the implementation. You do the thinking first.

This is the thing Briefli automates. It asks you the questions, you answer them conversationally, it produces the brief. The thinking still happens — it just happens faster, guided by the right questions.

Free at briefli.io →