A raw phone video can now become a captioned, edited short in about 15 seconds. A browser agent that cost dozens of dollars per run can be rebuilt to cost less than one. Production is collapsing in price. Almost nobody has repriced the document that tells production what to do.
I mean the brief. The one-page, sometimes one-paragraph, often one-Slack-message instruction that starts every piece of marketing work. For most of my career, it was treated as a formality before the expensive part began. That order has flipped.
When production was slow and costly, a vague brief got corrected along the way. The designer asked questions. The copywriter pushed back. The first draft came back wrong, and everyone in the review meeting quietly worked out what the brief should have said.
That friction was expensive, but it was also a safety net. Humans filled the gaps. A weak brief cost a few extra days and some frustration, and the work usually ended up roughly where it needed to be.
AI does not ask clarifying questions unless you build it to. It fills the gaps with the most statistically average answer. Give a model a vague brief and it will return fifty confident, polished, entirely generic variations, faster than you can read them.
Now the mistake ships at scale
This is the operational reality I see in agencies and in-house teams right now. Output volume has gone up five or ten times. The quality of the thinking behind it has not moved at all.
So a bad brief no longer produces one bad ad. It produces a hundred, and a testing framework that dutifully finds the least bad among them. The team reports a winner. Nobody notices that every option was built on the same wrong premise.
In engineering, we used to say that errors compound downstream. An error in the requirements costs far more to fix than an error in the code. Marketing just got its own version of that rule, and most teams have not noticed.
What a brief has to carry now
A brief that works with AI needs more than a product, an audience and a deadline. It needs the things a model cannot infer from the internet, because they exist only inside your company.
First, the one claim. Not five benefits, one thing you want the customer to believe after seeing the work, stated as a sentence they could repeat. If the team cannot agree on it, no amount of production will fix that.
Second, the proof. The specific data, customer quote or test result that makes the claim believable. Models will happily invent supporting evidence. Your brief should make that unnecessary by supplying the real thing.
Third, the boundaries. What the brand never says, which competitors you never mention, which tone is off limits, which promises legal will reject. These are the rules that a human team learned over years and a model learns only if you write them down.
Fourth, the test. How you will know the work succeeded, defined before anything is made. I argued earlier that evals are the new product spec. In marketing, the brief should carry its own evaluation criteria.
The brief is also the prompt
There is a practical reason this matters more every month. In many teams, the brief now goes directly into the tool. The same text a human once interpreted is pasted into Claude, ChatGPT or an agent workflow as the instruction.
That removes the last layer of translation. A strategist used to write the brief, and a creative lead turned it into direction. Now the brief is the direction, word for word, and every ambiguity becomes a decision the model makes on its own.
Teams that understand this start writing briefs differently. Shorter sentences, explicit constraints, real examples of good and bad output. It looks less like a marketing document and more like a specification, which is what it has become.
It also becomes reusable. A well-written brief for one campaign turns into a template, then into a standing instruction for the brand. Over time, that library of briefs is one of the few assets a competitor cannot copy by buying the same tools.
Who should write it
In many teams, briefs get delegated downward because they feel administrative. A junior account manager fills in a template, and the senior people show up at the review. That made a certain sense when the review was where the real decisions happened.
With AI, the review comes too late. By the time fifty variations land on the table, the decisions are already embedded in all of them. The senior judgment has to move upstream, into the document itself.
That is uncomfortable, because it exposes thinking. A vague brief protects everyone: if the work fails, the execution gets blamed. A precise brief puts the strategist's name on the premise. That accountability is exactly what makes it valuable.
I wrote recently that taste is what AI cannot copy. The brief is where taste becomes an instruction instead of a reaction.
A simple test for your next brief
Before you send any brief to a person or a model, hand it to someone outside the project and ask them three questions. What should the customer believe after this? Why would they believe it? What must this never say?
If they cannot answer all three from the page alone, the brief is not finished. It will not get finished in production anymore. It will just get multiplied.
Execution got cheap. Thinking did not. Spend accordingly.