What AI users are learning that every agency already knew
For most of my career, I’ve sat on both sides of the creative table. I’ve written briefs, rejected concepts and sat through third-round revisions. I’ve also been the person trying to work out what a stakeholder meant when they said “make it pop.”
When AI image and copy tools first arrived, a joke went round design studios. Designers didn’t need to worry about AI taking their jobs, because for that to happen the client would have to know what they wanted first. Jobs secure. 😂
It was funny because it was true. A few years on, it’s proving more accurate than anyone expected. Millions of people now type a request, get something back, and think: that’s not what I meant.
Welcome to our world. We’ve been dealing with this for decades.
The machine isn’t the bottleneck any more
A few years ago, the limit on AI was what it could do. Today it can write the copy, design the layout, build the website and draft the strategy, often in seconds. The constraint has moved. The hard part is no longer making the thing. It’s knowing what you actually want, and saying it clearly.
That is exactly the problem creative agencies have always had. The talent was rarely the issue. A good designer can execute almost anything. What sank projects was the brief: vague, contradictory, or missing the one detail everyone assumed was obvious.
“I’ll know it when I see it”
Every agency has heard this line. It’s the most expensive sentence in marketing. It means the client hasn’t decided what they want, so the creative team has to guess, again and again, until something sticks.
AI users now do the same thing at scale. They write a one-line prompt, react to the output, tweak, react again, and twenty iterations later wonder why it took longer than doing it themselves. The tool didn’t fail. The thinking that should have come before the request never happened.
The gap between what you said and what you meant
Agencies learned long ago that clients describe solutions when they should describe problems. “We need a video” is rarely the real need. “Young people don’t trust our brand” is.
AI has the same blind spot, only worse. A good account manager will push back and ask why. An AI will usually just make the video. It takes your words literally and fills every gap with the most average choice. That’s why so much AI output feels competent but bland.
What the people building with AI are saying
I recently watched Nate Herk’s interview with Matt Pocock, one of the most respected educators in AI-assisted software development. What struck me was how little of it was about tools or models.
Matt’s view is that AI has largely taken over the routine, tactical work. What separates good results from bad is the thinking around it: the planning, the structure and, above all, being clear about intent. He describes misalignment as one of the main ways work with AI agents goes wrong. The AI builds exactly what you asked for, which turns out not to be what you needed.
His fix is wonderfully low-tech. One of his best-known techniques simply gets the AI to interview you, relentlessly, before it starts any work. In other words, he makes the machine act like a good account manager and pull a proper brief out of you.
Any agency planner would recognise that immediately. The most valuable lesson in the age of AI turns out to be one the creative industry learned long ago.
Taste can’t be outsourced
AI makes options almost free. That makes judgement more valuable, not less. When anyone can generate fifty headlines, the skill that matters is recognising the one worth running. People who can tell good from merely plausible will pull ahead quickly.
What the agency world can teach AI users
- Write a proper brief. Who is this for? What should they think, feel or do? What does success look like?
- Give context, not just commands. The more an AI knows, the less it guesses.
- Show examples. “Like this, not like that” beats a page of adjectives.
- Describe the problem, not the solution. Let the tool suggest approaches you hadn’t considered.
- Let it question you. Ask the AI to interview you before it starts.
- Give specific feedback. “The tone is too formal for a 25-year-old audience” gets results; “I don’t like it” doesn’t.
The only skill worth having
Here’s the lesson I keep coming back to. The joke about clients never knowing what they want was never really about clients. It was about communication.
Whether you’re briefing a designer, managing a team, pitching to a board or prompting an AI agent, the outcome depends on the same thing: can you explain clearly what you want, why it matters and what good looks like? Tools will change every year. Models will get faster and cheaper. That skill won’t date.
So if you want to future-proof your career, don’t just learn the latest AI tool. Get better at communicating, with humans and machines alike. It’s the one skill that pays off in every scenario.
The brief was always the hard part. Now everyone has to write one.