From Paperclips & Frankenstacks to Cohesion: Highlights from the AI Edge Podcast

Marketing has had an unhelpful-assistant problem since long before AI showed up to try to fix it. Clippy used to pop into the corner of every Word document with unsolicited advice about letter formatting — a single, clunky assistant standing in for an entire toolkit. Marketers today don't have one clunky assistant anymore; they have a dozen good ones that don't talk to each other.

Marco Matos, co-founder and CEO of Adora, walked through what that actually looks like in a recent episode of The AI Edge Podcast, titled "The Frankenstack Is Killing Your Marketing — And AI-Native Is the Only Way Out."

The conversation ranged from Salesforce handing its brand equity to Claude to Nvidia's purchase of Hugging Face, but the most useful stretch was an honest look at why brand and performance teams keep landing on opposite sides of the same campaign. Let’s dive into the highlights.

Meet the Frankenstack

Every marketer Marco talks to is stitching together a creative tool, a measurement tool, and an agentic buying tool, none of which is anyone's actual job to run. His word for what results: a "frankenstack."

Every piece of that stack does its job well enough on its own. The trouble shows up in the seams. Campaign data lives in one tool, creative production happens somewhere else, and whoever runs buying and optimization is working from a third system entirely. Add a new point solution every time a new AI capability ships, and the stack doesn't get smarter. It gets heavier.

Finding Common Ground

For most of the industry's recent history, brand and performance have operated like rival departments sharing a building. Performance celebrates a campaign that drove clicks with creative the brand team never would have approved. Brand ships a beautifully produced asset with exactly one crop and one aspect ratio, leaving performance with nothing left to retarget against. Both sides are watching their own dashboards, working from their own definition of what counts as working, and the gap widens with every campaign.

Marco's read on the split is direct: "For years, it was always brand versus performance... they have to work together." The mechanism he described for making that real is what Adora calls Brand DNA: a model trained on a company's actual guidelines, product aesthetic, and campaign history, sitting underneath both creative production and performance reporting. The system flags which ads are working. That performance data feeds back into the same brand model that produced the creative in the first place, so both sides end up drawing from one shared picture of what's working.

The Ghost of Clippy Past

Clippy is the industry's favorite punchline, and for good reason. But Marco gave the paperclip more credit than it usually gets: "as a branding exercise, nailed it." His point: the interface was bad, but the system behind it, Microsoft's knowledge base and help center, "was amazing in what it provided later on."

That distinction lands differently now than it did in 2001. Marco and the host spent a chunk of the episode tracing how often the industry cycles back to a chat-based helper — ICQ and AOL Instant Messenger in the nineties, Clippy not long after, and now Salesforce handing over brand real estate to Claude while The Trade Desk rebuilds its interface around a conversational assistant called Zuma. As Marco put it, "it feels like it's all coming back, and we know that that didn't stick very long."

Marco's answer to what's different this time leaned on what sits behind the chat window, not the window itself: real first-party data, a company's actual brand guidelines, a system trained on what that specific brand looks like. His shorthand for how AI assistance gets framed these days is more co-pilot than replacement: "AI is putting the Iron Man suit around the human." Clippy never had that kind of power.

Practice & Perspective

Potential is high, but patience is not. Marco pointed out that it took the industry years to solve measurement for search and social, and by his own account, "in some respects, it hasn't been solved." Meanwhile, early data suggesting ChatGPT's ad units are underperforming is already circulating as a verdict, built on a test run with only a few hundred dollars of spend behind it.

It's been a matter of weeks, not years. Marco's own framing, in baseball terms: "I don't even think we've hit the Star-Spangled Banner yet. Batting practice is still happening on the field."

The rest of the industry moved just as fast in that same stretch. Salesforce and Anthropic launched ClaudeForce. Nvidia acquired Hugging Face. OpenAI shipped a new flagship model. The Trade Desk rebuilt its buying interface around a conversational assistant called Zuma. Google added an AI image editor across all of Workspace.

That's five separate shifts inside a few weeks. Judging any one AI channel's early performance against that backdrop skips over how much the ground itself is still moving.

About the AI Edge Podcast

The AI Edge Podcast covers what AI is actually changing about marketing and advertising, for the people who have to keep up with it on the job, not just at a keynote. Episode 34, featuring Marco Matos, is up on Apple Podcasts and Podbean.