Inside The Trade Desk’s Claude-powered campaign agent

According to one agency exec familiar with those conversations, a workflow built on Claude that connects to the platform can ingest a media plan, reformat it into a template that’s compatible with the ad tech vendor’s system and then build the campaign. From there, it can troubleshoot creative issues and generate recommendations on what to do next including the reasoning behind such suggestions. 

“We introduced Koa Agents as our effort to introduce agentic capabilities across our entire platform, everything from identifying and building audience strategies all the way to campaign setup, deployment and troubleshooting,” said Jordan Rost, vp of product marketing at The Trade Desk. 

That’s where it stops for now. The agent is not buying ads — at least not yet. Koa Agents are built on top of the existing Koa AI, the same system that has been embedded throughout its automated Kokai campaign management platform since it launched three years ago.

The agents add a new layer on top of that, focused not on what inside the platform once a campaign is live but on the work required to get it there  Rost said the longer-term vision is for agents to eventually touch the entirety of the campaign buying and management process, though autonomous bidding is not the near-term priority. For now, it is two different problems solved on two different points in the workflow, built on the same underlying AI stack.  

Speaking at the Marketecture Live conference in March, The Trade Desk CEO Jeff Green was asked directly whether someone could use Claude to create a campaign inside the platform. “If you’re a part of our closed beta, yes,” he said, before catching himself: “I wasn’t supposed to say that.”

The moment was unscripted but the thinking behind it wasn’t. Green has argued consistently that programmatic advertising is more naturally suited to agentic AI than almost any other industry. His reasoning: the sheer number of variables involved in running and scaling a campaign — bid factors, geo-targeting frequency caps, creative formats and channel mix — creates a decision-making headache that is genuinely hard for humans to navigate at speed and scale, and so therefore well-suited to machines. 

“When you take a $500,000 campaign and want to turn it into a million dollar campaign, you could change 10,000 different things,” Green said at the event. “One of them, or two of them, or some blend of them is better than another. If agents can frame these choices and help you see the trade-offs, this is a perfect task for AI.”

Koa Agents are the latest manifestation of that thesis. Crucially, the system is designed to work within a human-approved workflow rather than operate autonomously. It asks the marketer to sign-off before executing and explains its recommendations rather than simply making them. That emphasis on explainability is notable given the broader anxiety about AI systems that optimize for outcomes without showing their workings. 

Underpinning all of it is data. Rost was emphatic that the system is built not just to ingest data but to return it and therefore allowing advertisers to take campaign performance data out of its platform to train their own models. “We want our system to really respect and feed that desire for brand and agencies to get smarter in the process, ” he said. In an agentic world where first-party data is the primary competitive asset, that portability is either a meaningful differentiator or the price of entry, depending on the vantage point.

“The conundrum for The Trade Desk, and perhaps why they’ve been late to the game in this regard, is that simplifying programmatic buying is antithetical to their business model and their DNA,”  said Shiv Gupta, founder of staff-training service U of Digital. “They built their business by offering every knob and lever possible to traders to fine-tune campaigns and squeeze value out of them. This worked particularly well for agencies, who were then able to charge their customers for managing programmatic campaigns on The Trade Desk.”

That’s the kicker. Koa Agents might sound like a modest upgrade but the commercial logic is rooted in one of the fastest growing parts of the ad tech company’s business. The Trade Desk signed 45 joint business deals in March alone — a 55% jump on the same period a year ago – pulling advertisers into a direct relationship with a platform that has historically needed agency specialists to operate. Reducing that barrier makes obvious sense. What’s less obvious is the tension it creates. The Trade Desk built its reputation on giving agencies the tools to manage the complexity of programmatic. Automating more of that doesn’t just streamlining workflow, it starts to question whether that layer of expertise needs to exist at all. 

Rost framed it as democratization with agentic lowering the bar for advertisers and smaller agencies to build their own systems and connect them to The Trade Desk’s.

Which is what makes Koa Agents a genuinely interesting move. The Trade Desk is essentially testing a service that accelerates its attempt to automate away the complexity it built its business on, and hoping the thing it gets in return, stickier direct brand relationships, is worth more than the expertise economy it’s disrupting. 

Robert Webster, founder of AI marketing consultancy TAU, summed up the scale of that predicament: “Holdcos will build their own AI operating systems. Brands and independents can’t. MCP plus Claude gives them a planning layer that works across Google, Meta, TikTok and the rest without the build cost. That’s how The Trade Desk shores up the part of the market the holdcos don’t own.”