MadTech Daily: The Trade Desk’s Leaked Deck Revealing Its Agentic AI Plans; Nine Cutting the Value of Its Television Business

Full analysis

Three headlines in two minutes, and only one of them is actually news. The Trade Desk is building a conversational AI layer called "Ask Koa" that routes media buyers to specialized agents for audience creation, campaign setup, and troubleshooting. It comes from a leaked roadmap, not a disclosure, so treat the timeline as reported and unconfirmed. Alongside it: Nine Entertainment took a $403.6 million write-down on its TV business while its streaming service Stan grew 16%, and OpenAI put a shopping tab in the ChatGPT sidebar.

What's being decided (briefing frame): Nothing gets decided in this episode. It's a headline scan. The real question for an operator is which of these three is a durable structural signal and which is noise. The strategic read is a cheap Type 2 bet: watch, don't reorganize.

The Market Analyst Ask Koa is defensive product-building, and the market should read it that way. Every DSP is racing to put an AI agent between the buyer and the buying UI, because the UI is where switching costs live. If the workflow becomes "ask the agent," the agent becomes the moat, and The Trade Desk knows OpenAI-style natural-language buying is coming for that seat. Plain version: The Trade Desk is trying to own the conversation before someone else owns it. Nine's write-down is the loud one for investors, but it confirms what linear TV holders already priced in. The genuinely underrated line is OpenAI's shopping tab, a new lower-funnel surface that answers to nobody in the current retail media stack.

The Skeptic A leaked roadmap moving "alpha to closed beta" is a slide, not a product. Closed beta means it works for a handful of hand-held accounts under supervision. Agentic campaign setup that survives a real trading desk with real budgets and real edge cases is a different animal, and "the agent troubleshoots" is exactly the claim that dies first in production. As for OpenAI shopping: a tab is not a monetized ad surface. Until there's a paid placement auction, it's product discovery, and product discovery without a bid is just a feature. Don't reprice your roadmap off a sidebar icon.

The Operator Tuesday morning, none of this changes your job. Ask Koa isn't in your seat, and if it were in closed beta you'd be babysitting it, not trusting it. The thing to actually watch is the oversight model, because "the agent built the audience" is fine until the agent builds the wrong audience and spends real money doing it. Who reviews the agent's work, and how fast, is the operational question nobody in this episode touched. On OpenAI, the practical move is small and free: get someone watching whether ChatGPT shopping starts sending referral traffic, because that's the first tremor before any paid layer.

The Customer / End User (the advertiser) Advertisers didn't ask for a chat box on their DSP. They asked for lower CPMs and cleaner measurement. An agent that speeds up campaign setup is nice, but setup was never the bottleneck, outcomes were. The OpenAI shopping surface is more interesting to a brand than Ask Koa, because it's a potential new place to reach a buyer at the moment of intent. But brands should stay cold on it until OpenAI shows the economics. A discovery channel you can't bid into and can't measure is a curiosity, not a line item.

The tensions

Two real disagreements sit under the calm.

First, the Analyst and the Skeptic split on Ask Koa. The Analyst sees a land-grab for the buyer relationship that matters regardless of whether v1 is any good. The Skeptic says a closed-beta agent is a demo, and demos don't move budgets. Both are right on different clocks: the strategic intent is real now, the product reality is a year out.

Second, everyone splits on how to weight OpenAI's shopping tab. Is it the sleeper structural story, or a feature bar with no auction behind it? That hinges entirely on one thing nobody in this episode could answer: does OpenAI monetize it with paid placements.

What this actually hinges on

Ask Koa's importance hinges on whether agentic buying compresses the buyer's reliance on the DSP's own UI, which is where retention lives. That's the game, and The Trade Desk building it tells you the whole DSP category will build it too. The OpenAI story hinges on a single future event: a paid-placement mechanism. Until then it's a distribution surface, not a media channel. Nine's write-down hinges on nothing, because it's a confirmation, not a surprise. Linear repricing down is the base case, and this is another data point on a line already drawn.

The council leans: low immediate impact, one real structural thread worth tracking (agentic buying as the new DSP battleground), and one to monitor for a trigger (OpenAI monetization).

The prediction

The interesting call isn't about Ask Koa's ship date. It's that agentic buying becomes the defensive standard across DSPs fast, because none of them can afford to let a rival own the buyer's natural-language seat.

Prediction: By the end of Q1 2027 earnings season, at least two DSPs or SSPs beyond The Trade Desk (candidates: PubMatic, Viant, Magnite, StackAdapt) will have publicly launched or moved to open beta a conversational AI agent for campaign workflow, matching the Ask Koa concept.

Confidence: Medium. The competitive incentive is clear, but timing across independent roadmaps is hard to pin.

Why: The Trade Desk building a conversational layer that routes buyers to agents is a move to own the buyer's seat before natural-language buying erodes UI-based switching costs, and its rivals face the identical threat. Once one large DSP signals this is the interface direction, the others have to answer publicly or look behind, and "we have an AI agent" is cheap to announce even when the product is thin, which is exactly why several will announce. The opposite outcome, DSPs sitting quiet while The Trade Desk stakes the category, only happens if they judge the whole agentic-buying thesis wrong, and no public-company management team wants to explain to investors why they ceded the AI narrative.

Revisit by 2027-05-15: We're right if two or more named DSPs/SSPs publicly launch or open-beta a conversational campaign agent by Q1 2027 earnings season. We're wrong if The Trade Desk is still the only one with such a product shipped or in open beta.

One caveat on my own call: this leans toward the marketing incentive, and vendors announce adjacent AI products constantly, so the bar I'm holding myself to is a working conversational agent in open beta, not a press release with a roadmap slide. If all that shows up is more decks, grade it wrong.