AI Is Exposing The Creative Governance Most Companies Never Built

Every AI creative system runs on signals: it reads what audiences respond to, what performs, and what the last iteration taught it, then produces the next round against that read. When Meta and Google’s ad platforms began generating or altering creative that advertisers hadn’t approved, the failures exposed a governance gap that manual processes had concealed. Companies had long relied on experienced people to catch bad work by hand, often without defining who could approve it, which audience signals should guide the next iteration, or when the system had to stop. Agentic systems now produce and optimize creative faster than those informal checks can keep up. The faster the system runs, the harder that absence of formal judgment becomes to ignore.

Gregory Lewis Yancey spent years inside the marketing function at a large multinational bank before founding Lewis and Fields, where he now advises companies on AI strategy with a heavy emphasis on governance. He has seen the same problem across organizations ranging from global enterprises to growth-stage companies, and he keeps landing on the same diagnosis about where the real risk sits.

"When it comes to creative workflows, AI isn't really creating a governance problem. I think it's exposing the lack of governance that a lot of companies currently have," Lewis says. For him, that puts the focus on the people, approvals, and processes that were supposed to catch those problems and were never fully built.

Governance belongs at the start, not the cleanup

In Lewis' experience, governance enters the conversation only after something has already gone wrong, which he considers backward. When companies assess their readiness for AI, he wants governance among the first questions on the table, not the postmortem after a public stumble. Someone needs to own each decision. Teams need to define what an agent can do, what it cannot do, and who has to approve the work. And someone has to keep checking the output after it launches.

That last part is what teams most often forget. "They say, well, we have these guardrails and we have these frameworks in place, but then they forget about the continuous monitoring of that work. Guardrails aren’t set-and-forget. They require clear ownership, accountability, and continuous oversight. Moving fast without oversight creates risk," Lewis says. He is describing the launch-and-walk-away habit that can leave an iterative campaign drifting from the original guardrails while no one is checking the output. The difference between responsible speed and reckless speed comes down to naming who owns each decision and then watching the work as it runs in market.

Lewis puts the challenge on people. "The challenge isn't really the technology at all," he says, pointing instead to whether the organization has the controls, approvals, and accountability needed to decide which signals the system can act on. Review only works when the people running it have been deliberate about training the agents, closing the gaps that open when a team tells a system what to do and never tells it what not to do.

Training the system to speak the brand

Keeping a brand's identity intact while folding in AI returns Lewis to the old discipline of the brand playbook. Companies once trained creative teams on tone of voice, mission, and values, and he sees the same job now passing to the systems themselves. Good creative still starts with knowing the customer. AI changes how quickly that understanding can move through the creative process, turning it into hundreds of decisions and iterations before a traditional team would have finished one round. That makes the quality of what goes into the system much harder to ignore.

Oversight alone isn't enough, Lewis wants someone who knows the brand well enough to recognize when an otherwise acceptable output has gone wrong. "Every seasoned marketer knows that all great creative starts with a deep understanding of your customers and the insights that drive their behavior," he says. His years with large consumer companies taught him that nothing moved from brand positioning to finished creative without testing, and he sees AI making that habit more useful, because teams can now test more versions live and adjust them against real audience behavior.

Humans on both sides of the loop

The part Lewis thinks teams neglect is what happens on the consumer side. Brand judgment governs the signals going in; consumer testing tells the company whether the signal coming back is real. He wants a person accountable at each end, from the first insight through in-market optimization. Synthetic audiences have their place, but the final read has to come from actual people, because a machine testing against another machine can simply reinforce an assumption that was never verified. Large language models hallucinate, which makes that human check part of the process, not an afterthought.

Governing AI creative goes beyond controlling what the machine produces. It extends to the feedback loop that tells the machine what to produce next. The system generates work, reads the response, and uses that response to shape the next decision. A brand that never governs the quality of those signals can build something extremely efficient at making the wrong call, reading a false positive as a real trend and optimizing toward noise. The correction is the same discipline any signal-driven operator relies on: test the signal against genuine response before trusting it. Lewis describes the pipeline as two connected checkpoints, brand insight going in and consumer reality coming back, with a person standing at both.

Someone still has to say no

For Lewis, oversight only works if employees can actually override the machine. He wants someone with the authority to stop an asset when the numbers don't check out or the work doesn't sound like the brand, and he traces that authority back to training the people, not the technology. Too much attention goes to the tools and not enough to the humans running them. His answer is to make clear what the technology is doing, what the person is responsible for, and when that person has to intervene. Trust, he argues, matters more than speed, and giving someone the standing to catch a bad figure or a false brand note is how a company builds it with employees and customers alike.

The cost of getting that decision wrong is also moving beyond reputation. Under the EU AI Act, deployer disclosure rules took effect on August 2, 2026, requiring organizations to disclose AI-generated or manipulated image, audio, and video content, with penalties reaching €15 million or three percent of worldwide turnover. Weeks earlier, New York began requiring advertisers to conspicuously label ads with AI-generated human performers. "It's not just a brand reputational risk issue now. There's actual legal risk amplification," Lewis says. Work once judged on taste now carries compliance exposure, which raises the value of the person who can catch the problem before it ships.

That final call is the part Lewis will not hand off. The faster and more autonomous the system gets, the more consequential the decision to stand behind the work becomes. A machine can generate the asset without ever owning the choice to put a brand's name on it.