what AI taught me about briefing

i’ve been using AI long enough now that a pattern has become hard to ignore: when the output is bad, it’s usually my fault.

not always. sometimes the model just misses. but most of the time, when i look back at what i asked and what i got, the problem was upstream. the brief was vague, or it was too long, or i described the deliverable without describing the purpose, or i hadn’t actually decided what i wanted yet.

this is annoying to admit because it’s the same feedback i’ve been giving designers for years.

“what’s the goal here?” “who is this for?” “what decision should this help someone make?” the brief is the first design problem — get it wrong and everything after it inherits the confusion.

i knew this. i said it constantly. and then i started briefing AI the way most people brief designers — with a rough idea, a vague description, and an implicit expectation that the person on the other end would somehow fill in the gaps.

they don’t. they can’t. they produce something that matches the brief you wrote, not the brief you meant.

the difference with AI is that the feedback loop is fast enough to actually learn from. with a human designer, there are a lot of variables — maybe the brief was unclear, maybe they interpreted it generously, maybe they asked good questions in the brief, maybe they didn’t want to push back. it’s hard to isolate. with AI, the loop is tight. you write something vague, you get something vague back. you write something specific, you get something specific back. the correlation is almost too clean to miss.

so i’ve gotten better at briefing. more specific about purpose. clearer about what “done” looks like. more willing to spend time on the setup before i start on the output.

and then i went back and looked at some of the design briefs i’d written for humans over the years. they were not great. they were full of the same fuzziness that produces mediocre AI output — lots of description, not enough clarity on what success looked like or why the work mattered.

the AI didn’t make me a better briefer. it just made the gap visible fast enough that i had to do something about it.

i think that’s what a lot of AI tools are actually doing — they’re mirrors. not for the quality of the output, but for the quality of the input. they surface the thinking work that was always required and was always being quietly skipped.

whether you actually do that work is still on you.

the same applies to design work. if you want someone to engage with what you’re actually claiming — not just respond to what’s on the surface — i offer structured design critique. $75.