The machine scores, the human chooses
This was the week the argument stopped being theoretical. The AI jury, Google formats, Meta Advantage+ and Salesforce orchestration point the same way: AI is a real tool, but the call on what is good stays human.
Mariano De Vitto · June 2026
This was the week the argument stopped being theoretical. An AI jury trained on past Cannes winners was set loose on the 2026 Titanium shortlist, and it picked a different Grand Prix than the people did. In the same seven days, Google rebuilt the search ad around Gemini, Meta loaded Advantage+ with generative dubbing, music and imagery, TikTok pushed more of the campaign into automation, and Salesforce sent multi-agent orchestration into general use. The model that writes, scores and sorts is now wired into the daily work of marketing. The question is no longer whether to use it. The question is what we keep for ourselves.
Start with the honest case for the tools, because it is strong. A small team can now localize a proven video into ten markets in an afternoon. A search advertiser can answer a question instead of matching a keyword. A planner can buy connected TV with cleaner supply and harder measurement. A revenue team can stop letting leads go cold. These are not gimmicks. They remove real friction, and pretending otherwise is its own kind of dishonesty. The marketer who refuses to touch any of it will be slower and more expensive than the one who uses it well, and the client will notice.
The structural limit of the machine
Look at what the AI jury actually revealed. A model trained on what won before is, by design, a mirror of the past. It is brilliant at recognizing the shape of work that already succeeded and useless at feeling the jolt of work that has never existed. Titanium exists to reward the leap, and the machine voted for the echo. That is not a flaw to be patched in the next version. It is the structural limit of pattern matching. Optimization moves you toward the average of what worked. Creative leadership moves you somewhere the average has never been.
Three duties that do not delegate
This is why the order of operations matters more than the tools. Let the model generate the options, draft the variations, grade the entries, and surface what looks strong. Then put the decision in front of a person, because the decision is where context, taste and responsibility live. Someone has to decide what is true, since a model will produce a confident claim it cannot stand behind, which is why Google now forces disclaimers and Meta added a restricted-words list. Someone has to decide what is good, since taste is a read on a moment that training data always lags. And someone has to own the result, since an agent acting in your name does not answer to the customer when it gets the read wrong. You do.
None of this is a vote against the trend. It is a vote for keeping the human in the seat that actually requires one. The platforms will keep moving the line, quietly, one removed step at a time: the edit, the score, the cast, the follow-up, the choice. Most of those steps should move. One of them should not. The work that wins next year at Cannes will not be the work a model could have predicted. It will be the work a person had the judgment to make and the nerve to choose. Use the machine to widen the field. Keep the choosing for yourself.
The signal: Hand the machine the gathering, the drafting and the scoring. Keep three things human: what is true, what is good, and who answers for it.
The Signal Brief · Mariano De Vitto — Head of Marketing, Barcelona