Maya is a creative strategist at a mid-size agency. She’s launching a holiday campaign for Acme Outdoor across three channels: CTV, display, and social. She has one brief. She needs it running everywhere by Friday.
This walkthrough follows her journey through AdCP — from brief to delivery reporting — showing how the protocol connects creative tools, publishers, and AI agents into a single workflow.
Step 1: Write the brief
Maya starts with what she knows best: the creative direction.
In AdCP, the brief is the message field on build_creative. Maya doesn’t need to write JSON — her agency platform translates her direction into the protocol format:
acmeoutdoor.com/.well-known/brand.json. Every agent in the chain reads it from there. Maya’s team maintains it once.
Agency language → protocol terms
Agency language → protocol terms
Step 2: Discover what each seller supports
Before generating anything, Maya’s platform checks what formats each seller accepts. This happens automatically — but here’s what’s going on under the hood. Her platform callslist_creative_formats on each connected agent:
Now Maya’s platform knows: StreamHaus needs video files and VAST tags. Pinnacle needs display banners. CommHub needs feed-native content assets. One brief, three different output types.
Step 3: Check governance constraints
Before generating creatives, Maya’s platform checks what governance constraints apply. A buyer’s governance agent may require security scanning, content categorization, or policy compliance on all creatives. The platform callsget_adcp_capabilities on the buyer’s governance agent to discover what creative features will be evaluated:
registry:eu_ai_act_article_50), the creative agent ensures generated content includes required provenance metadata. If security scanning is required, generated creatives avoid patterns that trigger auto_redirect or credential_harvest flags.
Governance constraints are evaluated after generation via get_creative_features, which scores a specific creative manifest — but knowing the feature set before generation avoids a reject-regenerate cycle. Creative agents that are aware of governance requirements produce compliant output on the first pass.
See the Creative Governance overview for the feature-based evaluation model.
Step 4: Generate and preview
Maya’s platform routes the brief to the right creative agents. A CTV specialist handles video. A display agent handles banners. The social platform generates feed-native content that matches each community’s voice.
For CTV and display, Maya gets back draft-quality manifests she can preview immediately. For social, the platform will generate at serve time — but Maya can still preview what it would look like.
Maya asks to see the CTV spot in a living room context and the social post in two different communities:
build_creative call with an updated message. The agent iterates. This is the tissue session — fast, low-fidelity, focused on getting the direction right.
When Maya’s happy, she promotes to production quality:
Step 5: Distribute to sellers
Now the finished creatives need to reach each seller. Maya’s platform calls sync_creatives on each one — same creative, adapted to each seller’s required format:
Different sellers have different review processes. Pinnacle auto-approves based on brand safety rules. StreamHaus and CommHub do manual review. Maya’s platform tracks approval state per seller — a creative approved on one and pending on another is normal, not an error.
CommHub also checks community guidelines beyond standard ad policy. If a promoted post is rejected from a specific community, list_creatives shows the rejection_reason referencing that community’s rules.
Step 6: Campaign runs — AI generates variants
The campaign goes live. Here’s where it gets interesting. On StreamHaus, the CTV spot runs as-is — one creative, one variant (Tier 1). On Pinnacle, the display ads run with asset group optimization — the platform tests different headline and image combinations (Tier 2). On CommHub, the social platform generates promoted posts that match each community’s voice and trending topics (Tier 3). Maya doesn’t manage any of this. The protocol handles it. But she can see everything that happened.Step 7: Review delivery and variants
One week in, Maya pulls delivery data. Her platform calls get_creative_delivery on each agent and merges the results:
The social platform’s response includes engagement metrics in the ext field — upvotes, comments, shares — alongside standard delivery metrics. Display variants include the manifest showing which headline/image combination each variant used. CTV includes completion rates and quartile data.
Step 8: Replay what ran
Maya wants to see exactly what CommHub’s AI generated for the hiking community. She finds the top-performing variant in the delivery data and replays it:
Step 9: Pricing and billing
Creative agents can charge for their services. Pinnacle’s account with each creative agent includes a rate card — the agreed pricing. For the ad server, Maya’s platform callslist_creatives and each creative includes pricing_options. For the transformation agent, pricing appears on list_creative_formats instead — the format is the product. When build_creative runs, the response includes vendor_cost so the platform knows what each build cost.
After the campaign, Maya’s platform calls report_usage on each creative agent with the creative_id and pricing_option_id from the build response. This closes the billing loop — the creative agent can verify the reported cost matches its rate card.
Different agents price differently. The CTV ad server charges CPM (cost per thousand impressions served). The transformation agent charges per format adapted. The AI generation platform charges per image generated. The protocol handles all of these through the same pricing_options pattern — see the pricing specification for details.
The full picture
Every step uses a standard AdCP task. Every agent — creative tools, publishers, social platforms — speaks the same protocol. Maya writes one brief, and AdCP handles the rest: format adaptation, multi-agent routing, cross-seller distribution, variant tracking, unified reporting, and billing reconciliation.
Go deeper
- Key concepts: Assets, formats, manifests, and creative agents — the building blocks behind this walkthrough
- AI creative: AI creative for campaign teams — the strategic view for non-engineers
- Generative creative: Generative creative — Tier 1, 2, and 3 in depth
- Build an agent: Implementing creative agents — make your platform speak AdCP
- Orchestrate: Multi-agent orchestration — the engineering patterns behind Maya’s platform
- Get certified: The Buyer track teaches the full creative workflow through interactive modules