When the 73rd Cannes Lions opened on June 22, 2026, the jury finally said what it had dodged for two years: AI on its own no longer impresses anyone. The festival introduced its first Creative Brand Lion and a new AI Craft category that rewards only what human judgment and machine speed produce together, never AI alone. Juror Lili Jiang put it bluntly, AI can't replicate human vulnerability. The same week, Amazon used the Cannes stage to launch Alexa+ Agentic Ads and conversational display ads on the open web, Yahoo DSP kept embedding agents across media planning, activation and optimization, Google widened its hands free Ads Advisor and Analytics Advisor, and NBCUniversal demoed agentic orchestration across linear and streaming, live sports included. The jury told the room AI alone doesn't win. The platforms in the expo hall were still selling full autonomy. Nvidia, Palantir, Pinterest and Fox used the same festival to unveil infrastructure that lets agents talk to each other and run campaigns end to end, the industry isn't experimenting with agentic anymore, it's rebuilding its plumbing around it.

Why did Cannes reward craft over automation this year?

Because the jury stopped treating AI-only output as distinctive. The new categories only honor the result of human judgment and machine speed combined, not raw automation, a real reversal from two years of AI-forward celebration. The irony compounds right there on the same beach, Luma AI ran a separate competition offering $1 million to whichever team wins a Cannes Gold Lion using its tool, so while the jury stage said AI alone isn't enough, the industry floor was betting real money on exactly how far AI alone could go. Meanwhile IAB's 2026 outlook puts US ad spend growth at 9.5%, crediting part of that growth to accelerating agentic AI adoption. The industry is sending two contradictory signals at once, route the budget to automation, hand the trophy to the human.

Why does one mistake make a marketer stop trusting the tool entirely?

Behavioral science has a name for exactly this reflex, algorithm aversion. Dietvorst, Simmons and Massey's experiments found people abandon an algorithm far faster than they abandon a human forecaster who makes the identical error, even when the algorithm outperforms on average. An agentic tool that misdirects budget to the wrong audience once, or writes one line that's off brand, leaves a lasting suspicion regardless of its average hit rate. Yahoo DSP's own survey found 55% of marketers trust agentic AI to plan and execute tasks, and 20% openly say they don't. General Assembly's research found only 39% of teams are confident they're using AI to drive revenue effectively, with 46% still ambivalent. That ambivalence isn't noise. It's exactly what algorithm aversion predicts.

What does that distrust actually cost the budget?

Platform efficiency promises are modeled on the scenario where the tool runs unsupervised. But Ascend2's research found 54% of marketers fear automation will erode creativity and brand voice, which pushes them to manually edit output and re-check it before approval. Every hour of manual override means you're paying for full autonomy and receiving half autonomy in return, and that gap never shows up on the vendor's dashboard. Part of IAB's 9.5% growth figure is exactly this money, flowing into agentic tools. What that budget actually returns depends on how much control your team is really willing to hand over, not on the headline efficiency number in the sales deck. Put a number on it: if half of what you pay for an agentic license sits unused because your team keeps hand editing the output, that money isn't underperforming marketing spend, it's idle software capacity. No P&L line calls it out by name. It just shows up as a return that quietly falls short of what the deck promised.

• Measure your team's real override rate before buying the full autonomy tier. Don't benchmark against the vendor demo, track how often your own people manually correct the output over a month.
• Pilot the agentic tool on a bounded slice of budget first. Hand over one low risk segment, not the whole campaign, and let trust build at small scale.
• Design a review step that treats one error as a data point, not a verdict. Algorithm aversion is your team's natural reflex, a deliberately built check absorbs it instead of letting it kill adoption outright.

Buying autonomy isn't the same as buying trust. Cannes said so from the stage. Nobody says it in your budget meeting.

What this means for your next campaign budget

Walking away from agentic tools isn't the answer, IAB's growth number shows that money isn't flowing back the other way. What's practical is pricing your own team's actual trust level into the decision, not the vendor's headline efficiency claim. Cannes' jury said it plainly this year, the machine is fast, but the decision still belongs to a person. Build next quarter's budget around that, not around what the tool promised in the demo.