What Is Amazon Ads Agent? Amazon's AI Ad Platform

Ads Agent is Amazon's AI for advertisers, and since September 29, 2026, Amazon Ads Agent is also the name of Amazon's whole ad platform, which brings sponsored ads and Amazon DSP (demand-side platform) together. Its built-in conversational experience, in closed beta, plans, builds, optimizes and measures campaigns. Outside AI agents connect through the Amazon Ads MCP Server.

Until this week, "Ads Agent" meant a chat assistant inside the Amazon Ads console, and that is what most explainers still describe. So operators are asking two fair questions:

• Is this the chat we tried last year?
• Do we now use Amazon's agent, or keep running our own?

I read what Amazon has published about the rename to answer both.

The choice that matters is who decides the changes. Amazon's conversational experience proposes budget and bid changes from "your advertising goals and Amazon Ads signals", and you approve them.

Our agent works on the same platform through the Amazon Ads MCP, and the Budget Pacing Skill builds its budget proposal from the monthly spend target and seasonality our team set. The change waiting for approval is computed from our plan, and it is written only after we approve that exact proposal.

Amazon Ads Agent is now Amazon's whole ad platform, not a chat assistant

Amazon Ads Agent names the platform where your campaigns run. Amazon's September 29 announcement calls it "the new name of the unified platform, now powered by agentic AI", and the platform page groups it into three campaign types:

• Sponsored ads, priced at "cost-per-click (CPC) pricing with no minimum spend"
• Display, Video, and Audio (DVA+), which Amazon says begins rolling out in late October
• Full-Funnel Campaigns, which the announcement says are "now available to all advertisers in the United States"

The old consoles folded in. The platform FAQ says: "We brought sponsored ads and Amazon DSP together into one ad platform, Amazon Ads Agent. Your existing campaigns, reporting, and settings have carried over." DSP capabilities such as audience targeting, frequency management, deal IDs and custom bidding "all live in DVA+ with advanced settings."

What changed on Amazon's own pages

Amazon Agent Atlas holds the April version of Amazon's help page, "Understand Ads Agent", updated April 15, 2026. It said "Ads Agent is currently in beta", that it was "incorporated throughout your experience, such as in campaign manager and Amazon Marketing Cloud (AMC)", and that "There is no cost to use Ads Agent." Amazon rewrote that page on September 30:

Read the table as dated. Amazon says capabilities and access are "expanding over time", and the announcement says its new features "will roll out over the coming months."

The conversational experience plans, builds, optimizes, measures and supports, and waits for your approval

The conversational experience is the part of Amazon Ads Agent you talk to. Atlas retrieved Amazon's description of how it works: it "uses large language models combined with Amazon's advertising data to navigate and perform complex tasks", and "Ads Agent summarizes all proposed changes before taking action. You review and approve recommendations before Ads Agent makes any changes."

The September 30 help page keeps that control and lists the jobs by objective:

Atlas's April copy of the page names two of those jobs the same way: "Create campaigns from a media plan", which builds campaign and ad group structures from an uploaded plan, and "Optimize campaigns at scale", which updates budgets, bids, targeting and frequency caps in a conversational flow.

The example prompts are specific. Amazon's help page, in both versions, shows "For all campaigns with delivery rate below 80%, increase budgets by 25%" and "For all ad groups in campaign 'Q4-Sale', update the frequency cap to 1 per 3 days." Each one turns into a summary you approve.

In Amazon Marketing Cloud, Ads Agent drafts SQL but does not read your uploaded data

Ads Agent in AMC writes Amazon Marketing Cloud SQL from a plain-language question. Atlas retrieved Amazon's instructions: open the query editor and click "Generate SQL with AI", or click "Ask AI for AMC help" on the top navigation bar from any page in AMC. It generates audience and analytics queries, explains the query logic, and gives a side-by-side view to review changes before you run them.

Amazon is clear about two limits:

• Your uploaded data. The AMC help page says "the Ads Agent in AMC is not aware of advertiser uploaded datasets, and will not use them during SQL query generation."
• Your IDs. The best-practices page says it "doesn't look up lists of your campaign IDs, ASINs, keywords, or audience segment names", so you name them in the prompt "or modify the generated SQL before submitting the query."

That makes it a drafting aid for people who can read SQL, which is how Amazon positions it: its AMC help page says that to use these skills, "you should be familiar with" AMC privacy limitations, data aggregation thresholds and AMC's SQL syntax.

Our agent works AMC through the Amazon Marketing Cloud Skill, which reads the matching playbook before writing SQL, writes validated SQL, and returns privacy-safe answers or audiences ready to activate, all through the same Amazon Ads MCP connection that reads the campaigns. The AMC SQL introduction covers what that SQL is doing.

Outside agents reach Amazon Ads Agent through Amazon's MCP Server, with Amazon's Skills

The Amazon Ads MCP Server is Amazon's door for your own AI agents, built on the open Model Context Protocol. The platform page says you can "Connect your own AI agents and platforms to Amazon Ads Agent via the Amazon Ads MCP Server", and its FAQ says connected agents "can create, update, and optimize campaigns; run performance reports; manage audiences and creative."

Amazon's MCP overview names Claude, ChatGPT, Amazon Kiro, Amazon Bedrock and Amazon AgentCore among the applications that can connect. It also defines what travels with the connection:

• Skills: "a pre-defined prompt definition provided by Amazon Ads that guides AI applications through Amazon Ads workflows and tool usage."
• Expert tools: "Inference behind the tool interprets the request, applies Amazon intelligence, and produces the finished output." The AMC SQL Generator is the first, in closed beta.
• Hosting: "Amazon Ads builds, maintains, and updates the MCP Server."
• Access: Amazon's February 2026 launch post says the server and tools are "available globally in open beta to Amazon Ads partners with active API credentials." Its pages do not say whether an advertiser without API credentials can connect.

So from your own AI client, the server, its Skills and its expert inference are Amazon's, hosted and updated by Amazon.

Kuudo takes the other route: the Amazon Ads MCP runs in your own cloud, connects from any MCP client, and carries the Skills your team writes, beside the Selling Partner MCP for seller work in the same conversation. Amazon Agent Flow, the Amazon Agent Data layer, keeps the account history in your own storage. The Ads MCP Server comparison covers the two servers row by row.

Access works differently too. Each account consents once to Kuudo's managed Amazon Ads app, and every agent and tool you connect reuses that authorization. An agency brings its client accounts in without an Amazon Ads API app of its own.

What happens next: decide whose agent runs your Amazon Ads account

Your campaigns stay in Amazon Ads Agent whichever agent works on them, so the decision is whose agent does the work and whose rules it follows. Five steps settle it:

1. Check your access. The conversational experience is a closed beta by eligibility and locale; ask your Amazon Ads account executive.
2. Write down the rules your team applies. The margin a campaign must clear, the brand terms you defend, the pacing that fits your season.
3. Test Amazon's answers against those rules. Its recommendations start from your goals and Amazon Ads signals; see whether they land where your rules would.
4. List the AMC questions that need your own data. Amazon's agent will not use uploaded datasets when it writes SQL.
5. Give one agent the writes. One approval path and one record per set of campaigns.

If Amazon's answers already match how you run the account, its agent is built into the platform you use every day. If your rules, your data and your other Amazon work should shape every change, Kuudo vs Amazon Ads Agent puts the two side by side, row by row.

The rename makes the platform Amazon's by name, and the agent a choice. I will update this page as Amazon opens the beta.

Next: why an agent's change should wait for a person before an audience activates.