Agencies vs In-House: Who Wins with AI Agents?

Dentsu and Dept made their move. Both agencies are putting AI agents at the center of their go-to-market offerings, betting that brands will pay for agent-powered campaigns rather than build them internally.

Then Hyundai ran its own agentic media pilot. The brand went direct to the tech. No agency middleman required.

According to Digiday, this tension is exactly what agency leaders fear. The same AI agents they're pitching to clients are available to those clients directly. Hyundai proved you don't need an agency to run agent-based media. You just need the will and the workflow.

For SMB founders watching this unfold, the agency vs in-house debate just got more complicated.

The Agency Value Prop Has Shifted

Agencies used to win on execution. They had the teams, the tools, and the relationships. You hired them because you couldn't hire fifteen specialists yourself.

AI agents compress that advantage. An agency might deploy an agent that writes ad copy, another that optimizes bids, and a third that analyzes performance. But those agents are not custom-built for each client. They're the same agents you could license directly from the vendors building them.

The Dentsu and Dept approach is smart positioning. They're saying: we know how to orchestrate these agents. We have the prompts, the pipelines, and the human oversight to make them work. That's the new agency value prop. Not the labor. The configuration.

For an SMB, that's worth asking: is configuration alone worth the agency markup?

What Hyundai's Pilot Actually Proves

Hyundai's in-house experiment matters because it worked. They ran agentic media without an agency partner. The brand had the internal team to manage it, sure. But the actual agent work happened inside their own stack.

This tells you the technology is accessible. The moat around "we have AI tools" is gone. Vendors are selling these agents directly. The question becomes whether you have the operational bandwidth to point them at your GTM problems.

If you're an SMB founder with a lean marketing team, the Hyundai model might feel out of reach. Hyundai has scale. They have people whose job is to manage these systems. You probably don't.

But that gap is shrinking. The agents are getting easier to use. The interfaces are getting simpler. The Hyundai pilot is a preview, not an outlier.

The Real Decision Tree

Forget the agency sales pitch for a moment. Forget the vendor demos. Here's what actually matters when you're deciding between agency and in-house for AI-powered GTM.

Your team's capacity. Do you have someone who can spend ten hours a week tuning agent outputs? If not, an agency earns their fee by doing that work for you.

Your tolerance for iteration. AI agents need feedback loops. They produce drafts. You reject, refine, approve. If your internal team is already underwater, adding agent management to their plate creates a bottleneck.

Your data maturity. Agents perform better when they can access your historical performance data, customer lists, and creative assets. Agencies want this access. In-house teams already have it. The question is whether you can structure that data in a way agents can actually use.

Your growth stage. Early-stage SMBs often benefit from agency speed. You get a running start without building infrastructure. Mature SMBs with established GTM playbooks might find more value in owning the stack directly.

Where the Middle Ground Lives

This is not a binary choice. Some of the smartest SMBs I see are doing both.

They hire agencies for specific campaigns or launches. The agency brings agent-powered execution. The SMB gets results without the operational lift.

Simultaneously, they're building internal capability. Someone on the marketing team becomes the AI point person. They learn the tools, test the agents, and develop fluency. Over time, the agency relationship shifts from execution partner to strategic advisor.

This hybrid model hedges your bets. You get agency output now while building in-house muscle for later.

What to Ask Before You Decide

If an agency pitches you on their AI agent capabilities, ask three questions.

First, what exactly are you configuring that I couldn't configure myself? If the answer is vague, the value prop is weak.

Second, who owns the outputs and the data? If the agent-generated content lives in the agency's systems, you're locked in. That might be fine. But know what you're paying for.

Third, what happens when the agent makes a mistake? AI agents produce errors. They hallucinate, they misinterpret, they go off-brand. Who catches that? Who fixes it? If the agency says their agents don't make mistakes, they're lying.

The Bottom Line for SMBs

Dentsu and Dept are right to bet on AI agents. This is where GTM is going. But Hyundai is also right to test the in-house path. The technology is democratizing faster than agencies can build moats around it.

For most SMBs, the agency route makes sense right now. You get access to agent-powered GTM without the operational overhead. But that window is closing. The tools are getting easier. The vendors are going direct.

The agencies that win will be the ones who sell more than configuration. They'll sell outcomes, not just agent orchestration. They'll take responsibility for what the agents produce, not just point the agents at your problem.

The SMBs that win will be the ones who build internal fluency while using agency leverage. You don't need to pick a side. You need to know when each one serves you.

Hyundai showed what's possible. Dentsu and Dept showed what agencies are offering. Your job is to figure out which one gets you to revenue faster.

That's the decision. Not agency or in-house. Which path ships results.