AI's growth outpaces regulations, forcing marketers to adapt to a less controlled environment.
Security expert Robert E. Lee critiques AI blockers, noting their ineffectiveness against determined data scrapers.
Cost-saving measures in AI models can lead to degraded service, pushing marketers towards open-source solutions.
Lee predicts AI pricing will rise, likening it to Uber's initial low-cost strategy followed by price increases.
Marketers love control. Control over data; control over budgets; control over brand narrative. But AI doesn’t care about control, and the industry is colliding with that reality faster than anyone expected. As AI capabilities race ahead of regulations, marketing teams are finding themselves planning for a fantasy world instead of the realities of the one they actually operate in.
We spoke with Robert E. Lee, a product leader and most recently the Global Head of Intelligence and Creator Risk for TikTok Shop at ByteDance, who has spent over two decades on the front lines of digital security for platforms like Amazon, Twitter, and Snap. Over that time, he’s helped design and protect systems used by billions of people worldwide—the very platforms where marketers fight for attention, safeguard brand reputation, and manage sensitive customer data. Lee brings a candid perspective shaped by that scale of responsibility, and his take on AI’s growing role in marketing is anything but theoretical.
Cloudflare’s decision to block AI crawlers by default, for instance, sounded like a win for marketers intent on protecting data and content, but Lee is clear-eyed about its limits. "For anyone who is even a little bit technical, these kinds of blocks are not an insurmountable hurdle," he says. In practice, that means scrapers will still scrape, and any strategy built on the promise of perfect protection risks collapsing under real-world conditions.
The genie's out: "Encryption is essentially math, and you cannot legislate math. All it takes is one person to release an open-source version, and once that knowledge is out, you can't put the genie back in the bottle. Attempts to isolate or wall off content may be amusing or annoying, but they will never stop progress," says Lee. What that means in practice is simple: data will be scraped and content will be copied. The smart move isn’t betting on perfect protection but building the resilience to keep performing in spite of it.
But the illusion of control isn’t just about data access; it extends to the very AI products that marketers are increasingly relying on. Lee explains that the surprisingly low cost of models like GPT-5 hides a key trade-off: marketers and ad tech platforms lack true control over the quality and predictability of the service they pay for as these providers optimize for their own costs behind the scenes. The result is an ecosystem where campaign performance can silently degrade without notice, which can be a serious risk when budgets and brand trust are on the line.
Sausage-making secrets: "GPT-5 is not a single model; it's a router that directs your request to the cheapest possible model in its collection. OpenAI hides how the sausage is made, so you're effectively getting a degraded model whenever they think they can get away with it. That is how it's cheaper."
In response, savvy practitioners are already charting a different path. Instead of accepting downgrades from proprietary tools, they're building their own stack of open-source models. Lee argues these options often deliver more control and predictability for campaign optimization, measurement, and creative work. With the right setup, he said, marketing projects that once dragged on for weeks now wrap up in just a few days.
Weeks to days: "The cool thing about AI is that with the right tooling, projects that used to take two to three weeks are now taking two to three days. AI is one of those worlds that's at the same time overhyped and under-hyped. It's hyped for the wrong reasons," says Lee. The edge belongs to marketers who know how to choose the right tools.
A better backup: "As a heavy Anthropic user, when I experience timeouts on their top model, Opus, and they send me off to a lesser model like Sonnet, I switch to open-source alternatives like GLM-4.5 to finish my work. While not quite as powerful as Opus at its peak, these open-source models are superior to Anthropic's second-tier offerings."
Lee’s final warning is that today’s cheap AI won’t stay cheap. "We're going to see the Uber playbook repeat itself with the frontier models. Right now, these companies are giving us pricing reminiscent of Uber in 2016. Once they start charging at a level that provides them a real margin, it will be a price we won't want to pay," he says. For marketers, building an entire strategy on bargain APIs is a risky bet.
AI is already changing how campaigns are planned, optimized, and executed, but the real winners will be the ones who plan for volatility and build resilience into their stack. The edge belongs to teams that treat AI as a toolkit, not a safety net, and know how to pick the tools that actually deliver.
