Start with the verified scope. On September 29, 2026, Amazon Ads named its unified advertising platform Amazon Ads Agent and brought sponsored ads, Display, Video and Audio (DVA+), and Full-Funnel Campaigns into one AI-first operating environment. In the US open beta, Full-Funnel Campaigns lets an advertiser provide products, budget, and an optional ROAS threshold while the system allocates spend across sponsored ads, display, video, and streaming TV, selects formats, and continuously optimizes.
The change compresses more than time spent clicking through a console. It weakens a familiar agency value story: how many ad groups were managed, how often budgets were adjusted, and how many audience settings were touched.
ONCE’s view: agencies still matter when platforms can execute, but their durable value moves to defining what the system may do, detecting when it is wrong, taking control before risk scales, and turning outcomes into reusable judgment. Brands should procure that responsibility explicitly.
What Amazon Ads Agent can actually do
Amazon says Ads Agent can create campaign structures and ad groups from a media plan, adjust pacing, budgets, and delivery across many campaigns, recommend audiences, and help generate Amazon Marketing Cloud queries and audience segments through natural language. Full-Funnel Campaigns connects allocation, format selection, and continuous optimization to one objective. DVA+ spans streaming TV, online video, display, and audio, with a path between simplified automation and advanced settings.
“Agent” does not mean unlimited authority. Amazon also says advertisers can review recommendations, approve creative before launch, explicitly confirm DVA+ edits, pause campaigns immediately, and retain advanced controls over audiences, bidding, inventory, frequency, and creative strategy.
Those facts belong together. The machine’s executable scope is expanding while accountability remains with the advertiser. A human approval button is weak governance when the approver has no predefined standard for saying yes.
The market needs decision rights, not more confirmation boxes
IAB’s 2026 digital video buyer research found broad agreement that agentic AI will play a role in programmatic advertising, but no consensus on the role itself. Buyers called for humans in the loop, agent audit trails, and guardrails that limit what agents can do. IAB Tech Lab’s Agentic Advertising Management Protocols initiative is likewise building foundations for agentic systems to operate across advertising workflows using existing standards and taxonomies.
These efforts do not prove that one governance model has become the global standard. They do show the direction of the problem: once software can move real media money, prompt quality is secondary to authority, records, review, and accountability.
This is the agency opportunity. A brand should pay less for repetitive console labor and more for translating business objectives into executable constraints; detecting when platform proxies diverge from margin; scrutinizing inventory, frequency, and audience expansion; separating data delay from promotion effects, creative fatigue, or strategy failure; and preserving decisions that can be reviewed later.
The ONCE Media Agent Authority Sheet
Before enabling a platform agent or appointing an agency, put six fields into the same authority sheet:
| Field | Question that must be answered |
|---|---|
| Objective and anti-objective | Which business outcome may the system optimize, and which proxy metrics can never justify scaling by themselves? |
| Asset and inventory scope | Which SKUs, creatives, audiences, markets, placements, and publishers are allowed or excluded? |
| Financial authority | What are the per-action, daily, and flight ceilings, and which reallocations require escalation? |
| Action classes | Which actions may execute automatically, which may only become recommendations, and which are prohibited? |
| Evidence and audit | What prior state, rationale, approver, outcome, and rollback path must accompany every material change? |
| Stop and ownership | Who can pause when tracking, margin, compliance, or spend becomes unreliable, and what permits restart? |
The same sheet should govern the platform agent, the brand team, and the external agency. Its purpose is not to send every action to a person. It allows low-risk repetition to run while routing high-impact changes into approvals that contain enough information to matter.
Classify actions as green, amber, or red
Green: automatic execution
Within brand-approved objectives, assets, markets, inventory, and financial boundaries, an agent can handle routine pacing, allocation inside an existing campaign, and policy remediation under explicit rules. Thresholds should reflect the brand’s margin, sales cycle, and risk tolerance rather than a copied platform default.
Amber: recommendation with human confirmation
Cross-funnel budget shifts, audience expansion, higher frequency caps, new inventory, generated or replacement creative, destination changes, and conversion-event changes alter the business meaning of the campaign. The agent may calculate and recommend, but the brand or appointed agency should see the assumption, affected scope, and rollback path before approval.
Red: pause and recalibrate
Stop when conversion tracking fails; product price, inventory, or margin changes materially; unapproved claims appear; publisher or placement delivery cannot be explained; or the platform objective diverges from the company’s operating goal. Waiting for the algorithm to “learn its way back” is not a control.
Creative production belongs inside the same authority system
Amazon says Full-Funnel Campaigns and DVA+ can generate display, online-video, and streaming-TV assets from product pages, Brand Stores, or brand websites. Media and creative can no longer be treated as fully separate procurement packages. The buying system may generate an asset when a format is missing, or shift format allocation until a secondary asset carries substantial budget.
Connect the Responsive Commerce Creative BOM to the media authority sheet: which assets may be cropped, generated, or moved across markets, and which require human checks for product accuracy, sound, captions, and legal language. Media governance decides where money goes; creative governance decides what may carry it. They need the same approval state.
Change the agency’s unit of value
As platform agents spread, optimization counts and console hours become poor measures of agency value. Add an internal metric closer to operating judgment:
Cost per defensible intervention = (agency fee + governance and audit cost + exception-handling cost) ÷ material interventions with a recorded rationale, approval, outcome, and next action.
“Defensible” does not mean every intervention wins. It means the team can explain why it acted, what evidence supported the action, who owned it, and how the result changed the next decision. One correct pause or one rejected expansion can be worth more than dozens of mechanical bid changes.
A controlled starting point
Choose one product family in the US market and a test budget with a hard ceiling. Do not hand over the whole account. Complete the six-field authority sheet, retain an existing campaign or comparison condition, and name the approver for amber actions and the stop owner for red conditions. Review only the actions that materially changed budget, audience, inventory, or creative state.
Platform agents will not automatically erase agency value, but they will quickly erode services built around console activity. The new value lies in connecting business objectives, creative truth, media authority, and outcome learning into a decision system that can be audited, stopped, and reused.
Sources and evidence boundary
Verified facts come from Amazon Ads’ “Introducing Amazon Ads Agent, your AI-powered ad platform” and “Grow total sales with Full-Funnel Campaigns” (both September 29, 2026), Amazon’s current DVA+ guide, IAB’s July 14, 2026 Digital Video Ad Spend & Strategy summary, IAB Tech Lab’s AAMP project page updated September 22, 2026, and the NIST AI RMF governance playbook. Amazon’s early beta performance figures are not presented as industry averages. The six-field authority sheet, action classes, and cost formula are ONCE analysis.
