Agents of Change at Cannes Lions: Where Does a Human Stay in the Loop?

The Agents of Change interview series, recorded on the Croisette at Cannes Lions 2026. Produced in partnership with AgenticAdvertising.org. Sponsored by Agnitio.

We closed our last reel with a promise: after asking leaders where they already trust an agent today, and after asking where advertising's jobs are headed by 2036, there was one theme still ahead: where human judgment stays non-negotiable, no matter how far agentic advertising scales.

Nine leaders, one requirement

The prompt was simple: where do you require a human in the loop today? Nine leaders in nine different seats in the ecosystem. What came back, again and again, wasn't a checkbox. It was a description of exactly where accountability has to sit before an agent is allowed to act.

Ben Williams, Founder of Kiln, was candid that human oversight currently touches nearly everything he does as an agentic coder. His AdCP prototypes are human-in-the-loop by design right now. As they scale into production in 2027, he expects the reliance on manual review to shrink, not because oversight disappears, but because it gets engineered further upstream.

Gabriel DeWitt, Head of Publisher Platform Monetization, described Yahoo's default posture as multiple layers of human review stacked on top of each other: people evaluating incoming media plans and pricing, people reviewing submitted creative, people watching ad server pacing. The number of checks flexes depending on the deal — which is itself the point. The checks aren't uniform; they're calibrated.

Priti Ohri, CEO & Co-Founder, Advertible, put the boundary in structural terms: you don't hand your company or your IP to an agent. At Advertible, everything is blueprinted and architected by the engineering team first. The agent operates inside that architecture, not ahead of it.

Matic Tribuson, Chief Product Officer at Celtra, made the case that creative is where this may become the hardest to enforce. Without strong brand guidelines and review processes already in place, agentic execution doesn't create new problems — it just finds the cracks in old ones, and things can break even faster.

Why it has to be human

The second half of the conversation pushed past "where do you check agents" into "why does the check have to be human."

Chinmayi Bettadapur, VP and GM, Media & Entertainment at Salesforce, was blunt about where trust currently sits: fact-checking stays human, full stop. And for now, she noted, it's arguably agents that are in the loop on human work, more than the reverse.

Pamela Slea, CEO, Boltive, reframed the question as one of strategy, not tactics: it's not enough for teams to use AI tools to move faster. The human role is asking what the efficiency is actually for — whether it's making something genuinely better for the customer, not just faster to produce.

Mike Chevallier, SVP, Advanced Audio and Data Revenue at iHeartMedia, tied it directly to trust in the market: consumers are telling researchers they're tired of AI slop and want human interaction, even as the industry leans harder into agentic strategy. For him, that tension only resolves one way: with visible human governance behind every agentic workflow. Otherwise, "human-guaranteed" claims don't hold up.

Brian O'Kelley, CEO, Scope3, reframed the checks-and-balances problem entirely. The industry has always claimed a "four-eyes" standard on approvals, he pointed out, without always practicing it. Agents don't remove that discipline — rather, they make room for it by clearing away the rote double-checking so humans can spend their attention on the judgment calls that actually matter: which advertisers to work with, which risks are worth taking, what's genuinely on-brand. That's the part of the job, he argued, that will never be automated, because it's where values and ethics get debated, not just outcomes.

Benjamin Masse, President, AgenticAdvertising.org, closed the segment with the framing AAO has built its standard around. He described it as embedded human judgment: accountability that's designed in from the start, not something a human discovers after the fact. His analogy was blunt and clever: you don't want kids breaking a lamp in the living room and reporting it after; you design so they never get the chance to break it unsupervised. Critical thinking has to be built into the system on all sides, so no agent is ever running without someone accountable for what it does.

From Human in the Loop → to Embedded Human Judgment

In this interview, we found nine descriptions of the same structural requirement, from nine different vantage points. Williams' present review layer, DeWitt's stacked checks, Ohri's engineering-first architecture, Bettadapur's human-owned fact-check, Slea's strategic gut-check, Chevallier's visible governance, O'Kelley's attention on the judgment calls that actually matter, Masse's lamp-proofed living room — where the point isn't catching the mistake after it happens but designing the system so it can't happen unsupervised in the first place. These are decisions about data weighting, boundary rules, policy constraints, and more, made by the people who build and govern these systems, baked in before the agent ever runs autonomously.

That's Embedded Human Judgment (EHJ) — approval built into the process before an agent acts, logged and traceable, not discovered after the fact. Not a single gate at the end of a workflow, but human judgment distributed across the whole lifecycle of a system: the training data, the policy layer, the brief itself. What this episode makes clear is that the industry's leaders are already operating this way in practice.

What's coming next

We started this series asking where trust already lives, then asked what jobs look like a decade out. This episode is where those threads meet: the jobs that survive 2036 are the ones built around exactly this kind of judgment, while the leaders already running production agentic workflows today are the ones already practicing it.

Watch episode three: Agents of Change Podcast | Humans in the Loop: Embedded Human Judgment | AAO at Cannes Lions 2026 and see the full AAO at Cannes Lions playlist.

Go deeper: Read AAO's whitepaper, Building the Future of Marketing, and explore the Ad Context Protocol (AdCP) at agenticadvertising.org.

Join us: AAO membership is open to brands, agencies, publishers, and technology companies who want to shape the future of AI and advertising. Become a member today.