Agencies Face AI Agent In-Housing Threat

Digiday reported last week that agencies like Dentsu and Dept are building AI agents into their core offerings for clients. But there's a catch. Hyundai just ran an agentic media pilot specifically to test whether they could take that capability internal instead of paying an agency to run it. (Digiday)

The article frames this as an agency problem. It is. But for SMB founders, there's a different lesson here about what you're actually buying when you hire outside help.

Agencies sell you two things: labor and capability. Labor is execution. Capability is knowing what to execute, how to set it up, and how to iterate when it breaks. Most agency contracts blur these together. You pay a retainer. You get deliverables. You don't ask which part was the thinking and which part was the doing.

AI agents change that math. When the doing part gets automated, the capability part becomes visible. And that visibility cuts both ways.

For agencies, this is dangerous. Their margins live in the gap between what clients think execution costs and what it actually costs. AI compresses that gap. A media buying workflow that used to require three junior analysts now needs one senior strategist plus an agent stack. The client sees the same outputs. The agency keeps the savings. Until the client asks why they're still paying full freight.

For SMBs, this is an opportunity to get honest about what you need.

Ask yourself: are you hiring an agency because you lack the people to do the work, or because you lack the knowledge of what work to do? If it's the first, AI agents might let you bring that execution in-house sooner than you think. If it's the second, you still need external expertise. But you should negotiate for that expertise differently.

Here's what I tell founders who ask about agency relationships in an AI-native world.

First, audit your current vendors by capability, not output. Your PR firm sends you a weekly coverage report. That's output. The capability is knowing which journalists matter, how to pitch them, and how to recover when a story goes sideways. The report can be automated. The relationships and judgment cannot. Pay for the latter. Expect the former to get cheaper.

Second, pilot before you commit. Hyundai didn't hire an agency to run their agentic media test. They ran it themselves to see if they could. That's the right instinct. If you're considering a six-figure retainer for something an agent stack might handle, spend ten thousand dollars testing whether it actually can. Either you save yourself a bad contract, or you learn why the agency is worth it.

Third, negotiate for transparency, not just results. Agencies hate this. They want to sell you outcomes. But when outcomes depend on agent workflows that can be replicated, you need to know what's under the hood. Ask for documentation of the stack. Ask for the prompts. Ask what happens if you bring the capability internal. If they won't share, they're selling you mystery, not expertise.

The Digiday piece quotes agency executives who say they're not worried about in-housing because clients still need strategic guidance. They're half right. Clients do need strategic guidance. But the agencies that survive are the ones who charge for guidance and give away the execution tools, not the ones who bundle them together and pray the client doesn't notice the bundle is coming apart.

For SMBs, the practical takeaway is simple. The next time you review an agency contract, separate the thinking from the doing. The thinking is what you're paying for. The doing is what AI is eating. Adjust your budget accordingly.

Helix runs GTM automation for revenue-share clients. We see this dynamic play out constantly. Brands hire agencies because they don't have the bandwidth to run outbound, manage lead enrichment, or execute campaigns. They pay retainers for months before realizing the agency is just running a stack that could run anywhere. By then, they've overpaid for capability they could have owned.

Don't be that brand. Ask the hard questions early. Test the in-house hypothesis. And if you find yourself needing execution help, make sure you're paying for expertise that can't be automated, not labor that already has been.

The Hyundai pilot is a signal. The signal isn't that agencies are doomed. It's that the line between agency work and in-house work is being redrawn. Get on the right side of it.