# Worked Example: A Retailer Launches a Seasonal Line Across Six Countries **The setup.** Meridian is a mid-size apparel and home-goods retailer — roughly $900M in revenue, stores and e-commerce across the US, UK, Germany, France, Brazil, and South Korea. It is launching "Coastline," a summer-into-autumn lifestyle collection, in all six markets on the same day. A decade ago this would have meant a six-month agency engagement, a small army of freelancers, and a marketing department of forty. Today Meridian's marketing team is **five people**: a CMO, a brand director, a growth lead, a regional-markets manager, and a marketing ops engineer who manages the orchestration stack. They do not produce most of the work. They set the brief, hold the brand, judge taste, sign the legally and reputationally consequential decisions — and they orchestrate a fluid labor market of specialized agents that does the rest. Here is the campaign, start to finish. --- ## The walkthrough **1. The brief becomes a machine-readable spec (Day 0, morning).** The brand director writes a normal creative brief — the Coastline mood, the brand guardrails, the launch date, a $1.4M total budget, target ROAS by market. Meridian's **orchestrator agent** ingests it along with the brand's living context store (visual guidelines, tone-of-voice corpus, prior-campaign performance, the product catalog with SKUs and margins). It decomposes the brief into a dependency graph: strategy → creative concepting → asset production → localization → media planning → real-time buying → measurement. Each node becomes a work order with acceptance criteria. The orchestrator does not do the creative work; it dispatches, holds context, and recomposes results. **2. Funding the campaign as programmable treasury (Day 0).** The CMO authorizes an onchain campaign wallet seeded with **1.4M USDC**, with policy guardrails encoded directly: no single sub-contract above $40K without a human co-signature; a hard cap of $620K on media spend; a daily burn limit; an allowlist of agent identity classes that must be KYB-verified. Every agent that works on Coastline draws against this wallet within the guardrails. The treasury is the budget — there is no separate PO process, no invoicing lag. **3. Hiring the strategy agent (Day 0, afternoon).** The orchestrator queries an **agent marketplace** for a creative-strategy agent with proven apparel/lifestyle reputation and EU + APAC market fluency. It shortlists three by machine-readable metadata — reputation score, sample work, price, KYB status — and the growth lead picks one, *StratosBrand*, on the strength of its portfolio. The contract is an onchain smart contract: a fixed **$6,000** milestone fee, escrowed, released on the team's acceptance of a positioning deck and three campaign-platform directions. StratosBrand delivers in about ninety minutes; the brand director accepts direction two and edits the third into it; escrow releases. **4. The lead creative agent recursively sub-hires specialists (Days 1–2).** Meridian contracts a **lead creative agent**, *AtelierOne*, for a $22,000 milestone to produce the master creative system. This is where the labor market gets deep. AtelierOne does not generate everything itself — it runs its own harness and **sub-hires a roster of specialists**, each under its own agent-to-agent onchain contract paid out of AtelierOne's fee: - Three competing **image-generation style agents** (a film-grain naturalist, a high-saturation editorial look, a minimalist studio look), paid **~$0.004 per variant**. AtelierOne commissions ~2,400 hero-image variants across the three, an outlay of roughly $10. Nanopayments make this trivial; AtelierOne then scores them against the brand visual model and the StratosBrand direction and forwards a shortlist of forty. - A **copywriting agent** for headline and body systems (~$0.02 per variant, thousands of variants generated and ranked). - A **short-form video agent** that itself sub-hires a motion-design agent and a licensed-music-clearance agent — a third tier of recursion — producing 15s and 6s cuts at a **$1.10 per finished cut** consumption price. AtelierOne compares outputs across its sub-agents, recomposes a coherent system (hero look, type system, six video cuts, a variant library), and presents it to the team. The brand director kills the high-saturation look as "off-brand for autumn," keeps the naturalist direction, and approves. Milestone releases; AtelierOne settles its sub-agents automatically in the same transaction batch. **5. Localization across six markets with atomic FX (Day 2).** The regional-markets manager dispatches a **localization/transcreation agent** per market — not literal translation but cultural adaptation of copy, imagery, sizing conventions, and promotional framing. Korean and Brazilian transcreation get heavier human review (see below). These agents are operated by entities in different countries and want to be paid in their own currencies. The campaign wallet holds USDC; **atomic FX (StableFX-style)** settles each contract instantly — paying the Seoul agent in onchain KRW, the São Paulo agent in BRL, the Berlin and Paris agents in EUR — at the best available onchain rate, with conversion and payment as a single atomic event. The team never touches a foreign-exchange desk; currency is an abstraction layer. **6. Media planning and real-time buying (Days 3 onward).** A **media-planning agent** allocates the $620K media cap across channels and markets by projected ROAS. It then hands execution to a **real-time media-buying agent** that bids into programmatic ad auctions. Two things make this work economically: **nanopayments** let it pay per-impression and per-bid at sub-cent granularity in real-time clearing markets; and **tokenized card credentials over CPN** bridge to the legacy ad platforms (the big walled-garden DSPs) that still settle in fiat — the agent transacts natively in USDC and CPN delivers the last-mile value to each platform's billing endpoint. The buying agent runs continuous A/B and multivariate tests across the variant library, reallocating spend hour by hour. **7. Verified delivery and event-based settlement (launch + ongoing).** Media contracts release payment on **proving oracles**: a third-party verified-impression and viewability attestation, posted onchain, triggers escrow release for delivered inventory; a conversion-attestation oracle (tied to Meridian's e-commerce events) gates performance bonuses. No attestation, no release. A **measurement/analytics agent** continuously reconciles spend against verified delivery and ROAS, posts daily attestations to the campaign ledger, and flags underperforming creative for the buying agent to retire. **8. The campaign runs, measured and self-tuning.** Within the first 72 hours the buying agent has cycled through hundreds of creative variant combinations per market, concentrated spend on the winners, and trimmed the losers. The team watches a live dashboard, not a stack of weekly reports. Total elapsed time from brief to live, multi-market launch: a few days, most of it human deliberation over taste — the discovery, contracting, and settlement steps each happened in seconds. --- ## Where the humans sit **In the loop** (a human acts before work proceeds): the original brief and brand guardrails; selection of the strategy and lead-creative agents; acceptance of every creative milestone; the kill/keep calls on creative directions; sign-off on Korean and Brazilian transcreation; final brand-safety and ad-standards review before any asset goes live; co-signature on any sub-contract above the $40K policy threshold. **On the loop** (humans monitor and can intervene, but the machine runs): real-time media buying and bid optimization; per-variant generation and ranking; the thousands of nanopayment settlements; hour-by-hour budget reallocation; routine localization for the lower-sensitivity markets. The marketing ops engineer watches guardrail breaches and anomaly alerts; the growth lead watches ROAS and can pull spend. The five-person team spends almost none of its time *producing* and almost all of it *judging* — which is exactly where human marketing taste is irreplaceable. --- ## What runs onchain - **Identity:** every agent carries a cryptographically verifiable onchain identity, grounded in real-world **KYB/KYC** and carrying an **agent wallet plus credentials** (Circle Agent Stack). Accountability traces agent → wallet → KYC'd operator. - **Contracts:** org-to-agent (Meridian → StratosBrand, AtelierOne) and agent-to-agent (AtelierOne → its style/video/music sub-agents) arrangements are smart contracts with work specs, acceptance criteria, escrow, and event-based release. - **Payments and settlement:** USDC settlement; **atomic FX** for paying localization and regional media vendors in EUR, KRW, BRL; **x402** for the machine-to-machine payment calls; nanopayments for per-variant and per-impression spend; **CPN + tokenized cards** bridging to legacy ad platforms. - **Measurement attestations:** verified-impression, viewability, and conversion oracles post onchain, gating escrow release and performance bonuses. - **Treasury:** the campaign budget itself is a policy-governed onchain wallet. --- ## The agents hired | Agent role | How discovered | Pricing model | Settlement | How output verified | |---|---|---|---|---| | Creative-strategy (*StratosBrand*) | Marketplace shortlist, human pick | $6,000 milestone | USDC escrow, release on acceptance | Team accepts positioning + directions | | Lead creative (*AtelierOne*) | Marketplace, reputation-ranked | $22,000 milestone | USDC escrow; auto-settles sub-agents | Team accepts master creative system | | Image-style agents (×3) | Sub-hired by AtelierOne | ~$0.004 / variant | Nanopayments (USDC) | Scored vs. brand model; human shortlist | | Copywriting agent | Sub-hired by AtelierOne | ~$0.02 / variant | Nanopayments | Ranked vs. tone corpus; human edit | | Short-form video agent | Sub-hired by AtelierOne | $1.10 / finished cut | USDC; sub-hires motion + music agents | Team accepts six cuts | | Localization/transcreation (×6) | One per market, marketplace | Per-market fee | Atomic FX → EUR/KRW/BRL/USDC | Auto for low-risk; human for KR, BR | | Media-planning agent | Marketplace, ROAS track record | Fee + % of managed spend | USDC | Plan reviewed; capped by policy | | Real-time media-buying agent | Marketplace | Per-impression / per-bid | Nanopayments; CPN to ad platforms | Verified-impression oracle gates release | | Measurement/analytics agent | Marketplace | Consumption (per attestation) | USDC | Posts onchain attestations | --- ## What's still human or judgment-bound The machine generates, ranks, prices, buys, and settles. It does not own the brand. Several things stay firmly human: - **Brand taste and the "off-brand" call.** Killing the saturated look, reshaping a campaign line — these are aesthetic judgments rooted in a brand's identity and a market's mood. Agents can propose at near-zero cost and rank against a model of past taste; they cannot *decide* what the brand should become next. - **Sensitive creative and cultural calls.** Transcreation for Korea and Brazil got human review because cultural missteps are reputationally expensive and not fully captured by a metric. A human who knows the market still reads the room. - **Regulatory and ad-standards review.** Advertising claims, comparative-pricing rules, influencer-disclosure requirements, and platform content policies vary by country and carry legal liability. A human (with legal/compliance support) signs off before launch. The onchain trail makes accountability cleaner — but accountability is the point, not automation of the judgment itself. - **Strategic intent.** Why this collection, why these markets, what the campaign is *for* — that originates with the team. The agents are extraordinary labor; the purpose is human.