Automation does not remove creative governance. It makes the absence of governance much more expensive.
On September 29, 2026, Amazon Ads introduced Amazon Ads Agent as the AI-powered platform connecting sponsored ads, Display/Video/Audio and Full-Funnel Campaigns. The system can work from an advertiser’s objective, products, budget and creative, then coordinate formats and optimisation. Amazon also makes one operational fact explicit: advertisers approve the creative before launch.
That approval step is the strategic opening. When a media system can generate missing formats and move spend between streaming TV, online video, display and audio, the valuable deliverable is no longer a large folder of fixed exports. It is a controlled creative system that tells both people and machines what must remain true, what may change and who can release the result.
A platform review is not a product decision
Amazon’s current advertising requirements still place accuracy and rights obligations on the advertiser. AI-generated content must not mislead customers, claims may require substantiation, and advertisers must maintain the necessary permissions for music, images, talent and other elements. Amazon’s separate ad-review process checks creative, copy, product type and landing pages for policy compliance.
Those controls matter, but they cannot decide whether a render changed a product material, whether a feature shown is available in the target market, or whether a creator agreement covers a new media type. The brand must make those calls before scale turns a small mistake into a family of assets.
We previously examined how to protect the learning loop when AI rebuilds ads from spend signals. This is the earlier control point: deciding what an agent is allowed to rebuild at all.
The ONCE Creative Control Pack
Before creative enters an agentic campaign workflow, give every product family a six-part control pack:
- Objective. State whether the asset is meant to support awareness, traffic, sales, leads or new-customer growth. “Improve performance” is not precise enough because the objective changes how a system may allocate formats and budget.
- Truth lock. Record the model, appearance, material, feature, price condition, eligible customer and product action that cannot change. Connect important claims to approved product information.
- Rights lock. Mark talent, music, voice, location, third-party marks, territories, channels and term. Only assets cleared for paid media and adaptation should enter the variable pool.
- Variable layers. Define what may change: opening, sequence, crop, background, caption length, voice, end card or call to action. Give each layer a permitted range instead of saying “AI may optimise.”
- Approval gate. Name the owner of product accuracy, brand expression and final release. Separate changes eligible for batch approval from those requiring an individual preview.
- Measurement label. Preserve the objective, source master, changed variables, approver and launch date for every released version. Otherwise cross-format reporting cannot reliably inform the next production brief.
Choose automation by risk, not enthusiasm
A low-risk product with a current product page, simple claims, complete rights and changes limited to framing or pacing can reasonably start with AI-generated candidates followed by batch review. A regulated product, complex demonstration, market-specific specification, comparative claim or safety-critical action needs a locked master first and a much smaller variable surface.
Availability also matters. Full-Funnel Campaigns launched in open beta for eligible US advertisers that sell on Amazon. Amazon said DVA+ would begin a phased rollout later in October, while its current DVA+ guide describes limited US access and planned expansion. These announcements show Amazon’s direction; they do not prove identical access in every account or market.
Do not price this work by the number of generated files
The cost of an agent-ready creative system is:
product-truth audit + locked master production + variable modules + rights preparation + cross-format QA + approval operations + measurement labelling.
Divide that investment by the number of approved creative control units, not the number of files the platform can export. A unit counts only when its facts are correct, rights are clear, it serves at least one named objective-format combination, it has passed the required approval and it carries a stable label.
This protects both sides of the production relationship. The brand can compare suppliers on usable inventory rather than output volume. The production partner can charge for the judgment, clean source material and governance that make automation safe, instead of competing with the marginal cost of generation.
A practical starting brief for supply-chain brands
Choose one priority product. List five facts that cannot change, three creative elements that may vary and one owner for each approval type. Then request clean product plates, demonstration proof, text-free footage, separated dialogue and music, market-specific claim records, usage rights and stable asset IDs.
For brands working with Chinese manufacturing partners, this step is especially useful before packaging, firmware or regional specifications diverge. The agent should receive a current, market-approved source of truth—not whatever file happens to be easiest to upload.
Source note: Platform facts were checked against Amazon Ads’ September 29, 2026 Amazon Ads Agent announcement, its current DVA+ campaign guide, general advertising requirements and ad-review documentation. The Creative Control Pack and cost model are ONCE’s analysis, not an Amazon policy or a performance guarantee.
