TUNING NOTES · AI · MARKET
AI agents in CTV buying in 2026: what platforms describe
QUICK ANSWER
Streaming TV ad platforms describe AI in three different ways: automated optimization inside the platform, built-in conversational agents, and connections that let outside AI assistants run campaigns. Here is what each platform's pages say, and which controls to set.
By The Tuning Desk · CTV Ad Finder · Published 2026-10-17
What are the three kinds of AI in CTV platforms?
The first and oldest is optimization: machine learning that sets bids, pacing and audiences automatically. Most programmatic platforms describe some form of it. The second is a conversational agent inside the platform, where you type what you want in plain language and the platform drafts or makes the change. The third is a connection to outside AI assistants, so that a buyer can work from the assistant they already use, with the platform's data and actions available inside it.
What do platforms describe on their pages?
Vibe's MCP page says Vibe MCP connects Claude, ChatGPT or Gemini so that users can draft, launch and adjust streaming TV campaigns from the assistant, with OAuth sign-in and logged edits, and that the MCP and the Vibe Developer Platform API are included with every Vibe account. The page lists reads such as campaigns, creatives, audiences, metrics and inventory, and writes such as creating campaigns and editing targeting and budgets.
Source: Vibe MCP page
Amazon's page describes Amazon Ads Agent as a conversational experience to plan, create, optimize and measure campaigns in natural language, and says Amazon DSP is now part of Ads Agent.
Source: Amazon Ads: Amazon DSP
Other platforms describe AI optimization. The Trade Desk's page says its Kokai platform analyzes up to 15 million ad opportunities each second. MNTN states 1T+ behavioral signals per day and 4M bid decisions per second. tvScientific describes AI-powered optimization, and Tatari describes an AI-powered planning engine.
Source: The Trade Desk: our demand-side platform
Source: MNTN homepage
Source: tvScientific homepage
Scroll the table sideways
| Platform | Campaigns in plain language | AI or automated optimization | What the pages say |
|---|---|---|---|
| Amazon DSP (now part of Amazon Ads Agent) | Stated Amazon Ads Agent: a "conversational experience" to plan, create, optimize and measure campaigns in natural language. Source: Amazon Ads: Amazon DSP | Not on the pages we read | Amazon Ads Agent is described as a "conversational experience" to plan, create, optimize and measure campaigns in natural language. Source: Amazon Ads: Amazon DSP |
| Disney Campaign Manager (formerly Hulu Ad Manager) | Not on the pages we read | Not on the pages we read | Not stated on the FAQ we read. |
| MNTN | Not on the pages we read | Stated "1T+ behavioral signals per day" and "4M bid decisions per second". Source: MNTN homepage | MNTN states "1T+ behavioral signals per day" and "4M bid decisions per second" (MNTN's own figures). Source: MNTN homepage |
| Roku Ads Manager | Not on the pages we read | Not on the pages we read | An "AI video upscaler" for creative. AI optimization is not described on the page we read. Source: Roku Ads: how much does it cost to advertise on Roku |
| StackAdapt | Not on the pages we read | Not on the pages we read | AI optimization is not described on the pages we read. |
| Tatari | Not on the pages we read | Partly An AI-powered Planning Engine. | An AI-powered Planning Engine. |
| The Trade Desk | Not on the pages we read | Stated Kokai "analyzes up to 15 million ad opportunities each second". | Kokai, which The Trade Desk says "analyzes up to 15 million ad opportunities each second". |
| tvScientific (tvScientific by Pinterest) | Not on the pages we read | Stated "AI-powered optimization". Source: tvScientific homepage | "AI-powered optimization". Source: tvScientific homepage |
| Universal Ads | Not on the pages we read | Not on the pages we read | A creative tool that "turns your existing content into TV ads". AI optimization is not described on the page we read. Source: Universal Ads homepage |
| Vibe (vibe.co) | Stated Vibe MCP: draft, launch and adjust campaigns from Claude, ChatGPT or Gemini. Source: Vibe MCP page | Partly "AI Optimization" appears in the feature menu; details are not on the pages we read. Source: Vibe homepage | Vibe MCP connects Claude, ChatGPT or Gemini to "Draft, launch, and adjust streaming TV campaigns" from the assistant, over OAuth, with edits logged. The MCP page says "Vibe MCP and the Vibe Developer Platform API are included with every Vibe account". Source: Vibe MCP page |
Which controls should you set before an assistant touches budgets?
- Connect with the platform's own sign-in and permissions, not shared passwords.
- Start with read-only access for reporting.
- Require approval for every change to budget, bids or targeting.
- Check that every action is logged with who approved it.
- Keep hard spending limits in the platform that the assistant cannot change.
Where does this help, and where does it not?
Assistants help most with routine work: pulling a report across campaigns, comparing two periods, drafting a campaign for review, checking pacing, or setting up a test structure consistently. They do not replace the decisions that need judgement and accountability: how much to spend, which creative to run, how to design a holdout test and how to read the result. Treat an assistant's proposals as drafts, and keep a named owner for every live campaign.
What should you ask a platform about AI?
Ask which assistants or agents can connect, whether access costs extra, which data can be read, which actions can be taken, how actions are logged and reversed, and which security attestations cover the access. Ask for documentation rather than a demo, and test with a small campaign first. If AI-assisted buying matters to your team, use the finder's plain-language assistant answer to see which platforms' pages describe it.
How do agents change the work of a CTV team?
Less of a buyer's day goes on clicking through setup screens and pulling reports, and more on deciding what to test, reviewing proposals and reading results. That shift favours teams with a clear testing plan and clear rules about who approves what. It does not remove the need to understand the platform: an assistant can only work with the data and actions the platform exposes, and a buyer still needs to know what those mean.
It also changes how you evaluate platforms. If your team plans to work through an assistant, test the connection during the pilot: ask it for a report you already know the answer to, have it draft a campaign you then check by hand, and review the change log afterwards. A short, deliberate trial tells you more than any feature list about whether AI-assisted buying will save your team time.
Expect this area to change quickly. Platforms are adding connections and agent features through 2026, and the pages we will not stay current for long. Each profile on this site shows the date its AI facts were read; check the platform's own documentation for the latest before you plan around a feature.