In the space of five weeks, four of the largest advertising platforms on the open internet, some of them fierce rivals for the same media budgets, shipped what amounts to the same product. The Trade Desk called it Zuma. Yahoo called it the Agent Network. Amazon calls its version Ads Agent. Google calls its pair Ads Advisor and Analytics Advisor. Strip away the branding and every one of them does the same three things: it talks to the advertiser in plain language, it recommends or executes changes to a live campaign, and it explains its reasoning in something other than a spreadsheet. That is not a coincidence of product roadmaps. It is what happens when an entire industry decides, at the same moment, that the dashboard is the bottleneck.
The Pattern Nobody Missed This Month
The Trade Desk rolled out Kokai Zuma globally on August 27, its founder and chief executive Jeff Green framing it as the next stage of a platform he has spent three years building: “Kokai changed the way digital media is bought and sold programmatically, and Zuma builds on that foundation, making our platform faster, simpler and smarter.” Two months earlier, Yahoo DSP opened its Agent Network, an integration layer that lets 47 outside AI tools plug directly into campaign workflows. Amazon has spent the back half of 2026 pushing its Ads Agent, first built for Amazon Marketing Cloud, into the DSP itself. And on August 10, Google folded new agentic features into both Google Ads and Google Analytics, giving every advertiser a persistent AI layer sitting on top of the platform they already use.
None of these launched in response to a single competitor’s move. Each vendor describes its own agent as the product of a multi-year internal roadmap. That four separate roadmaps converged on the same shape in the same quarter is itself the story: the constraint they are all solving for is not a feature gap, it is the sheer operational load of running programmatic media at scale.
How Four Platforms Built the Same Copilot
The Trade Desk: Koa Inside Kokai
Zuma is not a new platform. It is a release layered onto Kokai, the interface The Trade Desk introduced in 2023 to replace its older Solimar system. The centerpiece is Koa, a conversational assistant that now handles campaign creation, audience discovery, frequency optimization and campaign troubleshooting through natural-language prompts, backed by a system the company says analyzes more than 20 million ad impressions per second and draws on data from more than 70% of the providers in its Data Marketplace. The company’s own release cites an average 32% improvement in CPA performance from Zuma’s upgraded modeling and forecasting, drawn from early adopters before the global rollout. Zuma also strips out the divisive “periodic table” campaign interface that defined the original Kokai launch, replacing it with a metrics-forward dashboard and bulk-edit tools that preview the impact of a change before it goes live.
Yahoo DSP: Opening the Agent Network
Yahoo took a different structural approach: instead of building every agent itself, it opened a framework for outside vendors to plug in. The Agent Network, announced June 18, connects advertisers to 47 partners across four workflow categories, audience and contextual targeting, campaign activation, creative operations and measurement, while keeping campaign execution and governance inside Yahoo DSP. “Agentic AI should make advertising simpler, not harder, and that starts with openness,” said Adam Roodman, general manager of Yahoo DSP, in the announcement. Partners quoted in the same release framed it as a coordination problem as much as a technology one. Kyle Cheasman, VP of business development at Innovid, said AI “will be most valuable when it can operate across the tools advertisers already use every day,” and Georgiana Haig, global strategy and partnerships director at MiQ, said an open framework “will help advertisers operationalize AI more effectively.” The bet is that no single vendor, not even Yahoo, can out-build the combined agent libraries of 47 specialists.
Amazon: Ads Agent Moves Into the DSP
Amazon’s Ads Agent launched in late 2025 inside Amazon Marketing Cloud, where it translated natural-language requests into SQL queries against retail data too complex for most marketers to query directly. Through 2026, Amazon extended the same agent into the DSP side of the business, letting advertisers upload a media plan as a spreadsheet, or select one from Amazon’s own Ads Planner, and have Ads Agent generate launch-ready campaign structures, budgets, pacing and targeting for approval before anything goes live. The direction is consistent with the rest of the market: collapse the distance between an advertiser’s stated intent and a running campaign, and keep a human approval step in between rather than full autonomy.
Google: Advisors for Ads and Analytics
Google’s version arrived through Ads Advisor and Analytics Advisor, expanded on August 10 with AI-generated overview cards summarizing account performance since an advertiser’s last login, a prompt box for generating custom insight on competitive shifts, and a Dashboards feature that turns a raw data export into a chart from a text description. Josh Moser, senior director of product management at Google, positioned the update as filling the gap between raw reporting and an actual decision. The clearest evidence of what advertisers actually do with it came from a customer quoted in Google’s own release: “Ask Advisor has become my go-to for directional checks on paid media performance,” said Kevin Marshall, paid media director at Gardyn. “That speed matters because it gives me time to dig deeper and make shifts that improve performance sooner.” Marshall’s framing is telling: the agent is not replacing his judgment, it is compressing the time before he gets to use it.
Why Now, and Why It Looks the Same Everywhere
Programmatic platforms have added automation for a decade, first in bidding, then in creative, then in audience targeting. What changed is not that AI arrived, it is that the interface finally caught up to how much automation is already running underneath it. A DSP account manager today is often responsible for hundreds of line items across dozens of clients, each with its own pacing curve, frequency cap and measurement setup. The dashboards built for that job assumed a human would check every number by hand. They do not scale to the account loads platforms now expect a single person to carry, which is the real reason The Trade Desk, Yahoo, Amazon and Google all landed on a conversational layer at roughly the same time: it was the cheapest way to make an already-automated backend legible to the person still accountable for it.
MarTech has covered pieces of this shift already this year, from ad platforms becoming analysts instead of dashboards to the industry’s parallel effort to standardize how buy-side and sell-side agents talk to each other at all, covered in Ad Tech Is Writing Two Rulebooks for AI Agents. This week’s rollouts are the retail front end of that same infrastructure push: the protocols determine whether agents can talk to each other across platforms, and the copilots determine whether a human ever needs to leave the chat window to get an answer.
The uniformity is also a competitive tell. None of the four platforms is claiming its agent is smarter than the others in any verifiable way; the public numbers that exist, The Trade Desk’s 32% CPA figure chief among them, are self-reported and drawn from early cohorts rather than independently audited. What each vendor is actually competing on is data access and integration depth: whose agent can see the most inventory, the most first-party signal, the most downstream conversion data before it has to guess. That is a different competition than the one advertisers are being sold, and it is worth remembering the next time a platform demo leads with the chat interface instead of the data behind it.
What It Means for the Marketing Leader
For a CMO or head of paid media, the immediate risk is not that these agents make bad decisions, it is that they make plausible-sounding decisions inside a chat window that is harder to audit than the tables it replaced. A recommendation to shift budget or expand an audience segment needs the same scrutiny a junior media buyer’s recommendation would get, not less scrutiny because a machine produced it faster. Teams adopting any of these four copilots should insist on the same thing from all of them: a visible reasoning trail, a preview of the change before it executes, and a log of what the agent actually did versus what it suggested. Amazon’s approval-before-launch step and The Trade Desk’s bulk-edit preview both point in that direction; whether Yahoo’s 47 third-party partners hold the same line will depend on each individual integration, which is exactly the kind of detail procurement teams should be asking about before they turn one on.
There is also a talent question underneath the product question. If Koa, Ads Agent and the Google advisors succeed at what their vendors say they do, compress the time between a question and an answer, the value of a media buyer shifts away from manually pulling reports and toward deciding which of the agent’s suggestions to actually act on. That is a harder skill to hire for than dashboard literacy, and most agencies and in-house teams have not yet rebuilt their training or their job descriptions around it.
How to Evaluate a DSP Copilot Before You Trust It
Three questions cut through the marketing on any of these launches. First, does the agent show its work, or does it just hand back a number? An agent that surfaces the segments, creative, and spend it is basing a recommendation on is auditable; one that only outputs a conclusion is not, no matter how confident the interface sounds. Second, what happens on approval versus autonomy? Amazon’s spreadsheet-to-campaign flow and The Trade Desk’s bulk-edit previews both keep a human in the loop before anything spends; that is worth confirming for any third-party agent plugged into Yahoo’s network, since governance quality will vary partner to partner. Third, whose data is the agent actually using? A recommendation is only as good as the signal underneath it, and platforms with deeper first-party and marketplace data, which is the real differentiator The Trade Desk, Amazon and Google are each quietly competing on, will produce more defensible answers than one guessing from campaign metadata alone. Ask each vendor to answer all three before the next renewal conversation, not after.
Source: The Trade Desk
