A UK agency automated design, then went back to human-led work
A UK agency CEO, a US fractional CMO and an Italian AI video founder on what machines now do in creative work, and the part nobody has handed over.
A UK agency automated its design work. Then it went back to human-led work.
A US CMO over-automated creative review. A product launch got delayed.
An Italian founder builds AI that scores video clips. His customers still want a person on the final cut.
Creative work is where AI looks most finished. It is also where the gaps show.
We asked which parts of a job AI has actually taken, and which still need a person. Three people in creative work answered, and none said all or nothing. It matters now because every team is deciding what to hand the agent.
The scoring went to the machine. The taste didn't.
Alessio Battagliero, Founder & CEO of GPT-Video (Italy)
“GPT-Video can transcribe a long podcast, interview or stream, score candidate moments from 0 to 100 with reasons (hook, payoff, clarity, quotability), reframe to 9:16 with face tracking and burn in captions.”
“Customers still insist on keeping a human for taste and the final cut.”
Read the verbs in the first quote: transcribe, score, reframe, burn in. Every one is a step. The thing customers keep is not a step.
Fully automated. Then walked back.
Dennis Beytekin, CEO of Scale Oxford (Oxford, United Kingdom)
“We'd often spend so much time correcting small mistakes that the automation wasn't saving us much time anymore.”
“Small mistakes, generic AI aesthetics, and limited creative control eventually pushed us back to human-led design assisted by AI.”
The automation had a price: people kept fixing its work. The AI stayed on, but a person took the wheel back. That is not an anti-AI story. It is a story about who holds the pen.
The review stage that cost a launch.
Haiko de Poel Jr., fractional CMO of Mass Impact (Ponte Vedra Beach, United States)
“Over-automating the creative review stage once delayed a product launch because the system could not catch subtle positioning mismatches that only experienced judgment spotted.”
“Both proves strongest when AI manages production volume while a person steers the overall narrative and compliance needs.”
Production volume is where AI earns its keep. Review is where over-automation hurt. The failure was quiet: a subtle mismatch slipped through, and only experience caught it.
Where they split
Dennis Beytekin automated design, then moved back to human-led work with AI assisting. Small mistakes, generic aesthetics and limited creative control were his reasons.
Haiko de Poel Jr. keeps AI on production, like first layouts and basic element placement, and a person on narrative, positioning and review.
Alessio Battagliero's product does the scoring and reframing. His customers keep "taste and the final cut."
All three put the line in the same place: AI takes steps, people keep judgment.
They just found it differently. One walked back a fully automated process, one lost time on a launch, one hears it from customers.
My take
AI is not bad at creative work. If a role has a production half and a judgment half, let the agent take production and keep a person on the judgment. Then look hard at the review step, because that is where the quiet mismatches get through.
Before you automate anything, ask one question: if this goes wrong quietly, who notices?
Your turn
Where does your team draw the line between what the agent makes and who signs off, and what did it cost to find that line?
Next: AI drafting replies for about 700 storefronts with a person approving each one, a founder answering his own help desk, and an expired price and a frustrating bank bot as warnings.
