Creative Agencies Are Using AI Wrong

A year ago, creative agencies were navigating client contracts with strict “no AI” clauses. Today, those same clients are asking a different question in RFPs: “How are you using AI to drive efficiency, innovation, and value?” Most agencies are not ready for what that question actually means.

I have been thinking about this question because of a conversation shift I keep hearing about. The move from defensive positioning—proving you are not using AI irresponsibly—to proactive expectation—demonstrating how AI creates advantage—signals something important.

It signals that AI has moved from risk to expectation. But here is what interests me: I am not sure the agencies are ready for that shift. Or more precisely, I am not sure they are thinking about AI in the right way.

The Speed Trap

Most of the conversation around AI in creative agencies has centered on velocity. Digitas embedded AI agents into brainstorming and strategy sessions, cutting three days off tight brief timelines. VML uses WPP Open to create storyboards and creative variations at scale. These are real gains. Time matters in advertising.

But focusing solely on speed leads to what Jameson Proctor, CEO of Athletics, calls “the AI productivity trap”—working faster without necessarily working smarter, or better. And I think that is exactly where most agencies are getting stuck.

Here is the pattern I see: Agencies adopt AI tools. They generate more variations, explore more directions, move through rounds faster. Clients are happy with the speed. But nobody is building systems to capture what actually works.

Teams can produce a hundred variations in an afternoon, but they cannot systematically answer: Which approach resonated with the audience? Why did that concept perform better? What can we learn that applies to the next brief?

So they are not really getting smarter. They are just repeating effort faster.

What Clients Actually Want

There is a tension here that I think agencies are missing. When clients ask about AI efficiency, they are not just asking about turnaround time. They are asking: Are you using this technology to develop proprietary insight? Are you building knowledge that makes you more valuable over time?

Think about what VML's Brian Yamada said about their Virgin Voyages campaign, where users created personalized video invites voiced by Jennifer Lopez. It generated 25,000 invites and 2 billion impressions. That is impressive.

But here is the more interesting question: What did VML learn from those 25,000 variations that they can apply to the next campaign? Did they capture which message structures worked? Which visual approaches drove engagement? Which personalization elements mattered most?

If the answer is “we generated 25,000 things and moved on,” then speed is not creating competitive advantage. It is just creating more work to do again next time.

The Knowledge Capture Problem

The issue is not that agencies are not using AI. It is that they are using AI for outputs, not for learning.

Every AI generation contains implicit knowledge: what prompt worked, what model settings produced the best results, which iterations resonated in client reviews, which approaches performed in market. That knowledge is institutional gold—the accumulated intelligence that should make an agency more effective over time.

Right now, that knowledge evaporates. Prompts get lost. Successful workflows cannot be replicated. The learning curve resets with every new project.

This is the opposite of how expertise is supposed to compound. And it is why speed alone does not answer the client's real question. Clients are not asking “can you move faster?” They are asking “are you learning?”

The Agency Knowledge Gap

Here is the uncomfortable truth: most agencies have no infrastructure to capture AI-enabled learning.

They track billable hours. They archive final deliverables. But the process knowledge—the prompts, iterations, A/B insights, model configurations that produced the best results—lives in Slack threads, Discord servers, and individual memory.

When a creative director leaves, their AI expertise leaves with them. When a team transitions off an account, the institutional knowledge of what worked does not transfer. Every project starts with the same discovery process because there is no organizational memory of past discoveries.

Speed without memory is not efficiency. It is expensive repetition at higher velocity.

What “Using AI Right” Actually Looks Like

The agencies that will win the next phase are not the ones generating fastest. They are the ones building systems to learn from what they generate.

This means:

Systematic capture: Every AI-enabled project feeds a knowledge base of what worked, what did not, and why. Prompts, parameters, performance data—all preserved and searchable.

Institutional learning: When one team discovers an effective approach, that knowledge transfers to other teams. The organization gets smarter, not just individual practitioners.

Client-specific intelligence: Over time, you build a proprietary understanding of what works for each client—not generic best practices, but specific patterns that predict success for their brand, their audience, their objectives.

Provenance documentation: When clients ask (and they will) what was AI-generated and how, you can show them. Not as a defensive measure, but as proof of rigorous process.

The agencies that build this infrastructure will compound their advantage with every project. The ones stuck in the speed trap will keep running faster in place.