Every studio has now run the experiment: let AI write the copy, generate the concepts, draft the code. The results split cleanly. Teams that treated the output as a first draft got faster and arguably better. Teams that treated it as a final product shipped work that smells machine-made — and clients are developing a nose for it.
What AI is genuinely great at
The honest wins are real and unglamorous. First drafts that break the blank page. Twenty headline variations in the time it took to write one. Summarizing a fifty-page brief before a kickoff call. Boilerplate code, alt text passes, meeting notes. Anything where volume and speed matter more than voice.
Where judgment stays human
- Taste — AI averages what exists; positioning requires choosing what everyone else wouldn't
- Context — it doesn't know your client's history, politics, or the thing the CEO hates
- Accountability — when work is wrong, "the model did it" is not a sentence you can say to a client
The workflow that works
Generate wide, select narrow, refine by hand. Use AI to explore ten directions cheaply, pick the one with promise, then apply full human craft. The economics change — exploration becomes nearly free — but the standard for what ships doesn't move an inch.
The tell
You can spot AI-slop portfolios instantly: fluent, generic, interchangeable. That sameness is the ceiling of unedited output. The differentiator in an AI-saturated market is, ironically, the thing it can't generate — a point of view.