Agentic Commerce and the Advertising Layer

The agentic commerce moment

AI agents are becoming the front door to commerce. Consumers are using agents to research products, compare options, negotiate prices, and complete transactions - and the numbers are no longer projections. They're market data.

McKinsey projects up to $1 trillion in U.S. B2C agentic commerce by 2030. Morgan Stanley estimates $190-385 billion in U.S. e-commerce via agentic shoppers in the same timeframe. Bain reports 30-45% of U.S. consumers already use generative AI for product research, and AI accounts for up to 25% of referral traffic for some retailers today.

The shift is visible at the company level. Fortune reported in March 2026 that some brands now attribute 10% of revenue to agentic channels. Target's traffic from ChatGPT grew 40% month-over-month. Gartner projects 25% of enterprise software purchases will be mediated by AI agents by end of 2026, and 33% of all applications will feature agentic AI by 2028 (up from less than 1% in 2024).

The three-layer stack

The infrastructure for agentic commerce is forming in three distinct layers. Protocols and platforms are being built by the largest companies in tech. The advertising layer - how brands get discovered and recommended inside agent conversations - is fragmented and early.

The protocols exist. The platforms exist. The advertising layer doesn't. That's the structural gap. The same gap that created AdSense for the open web, AdWords for search, and offer walls for mobile games exists in agentic commerce right now.

How brands get discovered by agents

The major platforms are building discovery mechanisms for their own ecosystems. None of them serve the open agent ecosystem where independent developers build and deploy agents.

Google launched Universal Commerce Protocol (UCP) in January 2026, co-developed with Shopify, Etsy, Wayfair, Target, Mastercard, Visa, and Stripe. Direct Offers ads are in alpha testing inside AI Mode. Google is also introducing Business Agents for retailers - Lowe's, Michael's, Poshmark, and Reebok are already live.

Microsoft launched Copilot Checkout in January 2026. Merchants can close sales directly inside Copilot conversations. Brand Agents are live for Shopify merchants. PayPal and Stripe integrated.

OpenAI launched ChatGPT ads in February 2026 at $60 CPM. Working with Omnicom, WPP, Target, and Adobe. $250K minimum spend. Hit $100M ARR in 6 weeks with 600+ advertisers. Also co-developed ACP with Stripe for agent-to-business transactions, and signed Smartly as their first creative ad-tech partner.

CJ Affiliate launched an AI Visibility and Optimization solution that measures brand presence across ChatGPT, Gemini, Perplexity, Claude, and Copilot - and optimizes affiliate strategies based on LLM visibility gaps.

HBR published "Preparing Your Brand for Agentic AI" in March 2026. Their core framework: brands need to solve for discoverability (being found by agents) and desirability (being chosen once found). BCG published a similar piece identifying specific agentic scenarios marketers must prepare for.

Every one of these solutions is platform-specific or walled-garden. Independent agents built on ElizaOS, LangChain, CrewAI, and other open frameworks have no equivalent discovery infrastructure.

Walled gardens vs. the open network

Today's agentic advertising options are either expensive walled gardens or early-stage alpha products. None serve the open agent ecosystem at accessible price points.

This pattern has played out before. The early web had AOL and CompuServe - walled gardens that controlled content, commerce, and advertising within their walls. The open web won. Early mobile had carrier decks that curated app discovery. App stores and open ad networks (AdMob, Unity) won. In both cases, the long tail of independent publishers created more inventory than any single platform, and the open exchange captured the economics.

Agent commerce is following the same trajectory. ChatGPT, Copilot, and Gemini are the early walled gardens. But the open ecosystem of independent agents - built on ElizaOS, LangChain, CrewAI, and custom stacks - will produce more conversational surface area than any single platform. The open advertising layer will capture that.

The competitive landscape

Several startups are building ad infrastructure for AI agents. They differ in insertion model, revenue type, and which part of the agent ecosystem they target.

The critical gap: no competitor unifies all revenue types into one auction. ZeroClick does CPC. Kontext does CPM. ChatAds does affiliate. A quality-weighted unified auction accepts bids from CPM advertisers, CPC advertisers, affiliate partners, and lead-gen buyers simultaneously. All bids are normalized into a single score. The highest-scoring placement wins regardless of revenue model.

This is the AdSense play: more bidders competing for every impression means higher clearing prices and better publisher economics.

Unit economics

The economics of AI agent advertising hinge on two numbers: inference cost per task and revenue per ad impression. Both are moving in the right direction. Inference costs are declining roughly 10x per year - GPT-4-equivalent inference dropped from $20 per million tokens in 2022 to $0.40 today.

Inference cost per task (March 2026)

Revenue per ad impression

The unified auction thesis

When all ad types compete in one auction, publisher revenue increases because more bidders compete for every impression. A conservative blended eCPM model - combining display (60% of impressions at $10), CPC (20% at $8), affiliate (15% at $600), and lead gen (5% at $1,000) - yields approximately $147 blended eCPM. That's 2.5x OpenAI's $60 CPM for ChatGPT ads.

With 3 slots per conversation, that's $0.44 per conversation - enough to cover the inference cost of virtually any agentic task. Simple Q&A tasks ($0.01-0.03) are already coverable with display CPM alone. Medium tasks ($0.05-0.15) are viable with CPC in high-intent verticals. Complex tasks ($0.80-2.50) require affiliate, commerce, or lead-gen revenue - or serve as a partial subsidy where ads cover 30-40% and the user pays the rest.

Inference costs are declining roughly 10x annually. By 2028, even complex agentic tasks may cost under $0.02 - making virtually every conversation ad-coverable.

High-value verticals

Not all agent conversations are equal. The highest-value verticals are those where conversational recommendations drive purchase decisions with high per-conversion payouts - enough to cover inference costs today, even for complex multi-step tasks.

"Tasks covered per conversion" shows how many agent interactions a single conversion can fund. A $150 insurance lead covers 100-300 simple agent tasks or 1-3 complex agentic tasks at today's inference costs. A $500 financial services lead covers up to 1,000 tasks. These are the verticals where free, ad-supported agents become economically viable first.

The pattern

Every content layer in internet history went free. And every time it did, a monetization layer emerged that became worth more than the content itself.

The mechanics are identical every time. A new content surface emerges. It starts behind a paywall or subscription. Competition drives it toward free. The monetization layer - the network that matches demand to attention - captures the economics of the shift. DoubleClick did it for web pages. AdWords did it for search. Facebook Ads did it for social. AdMob and offer walls did it for mobile games.

AI agents are the next content layer. 61% of U.S. adults have used AI, but only 3% pay for premium. ChatGPT has 800M+ weekly users, 95% on free tiers. Inference costs are declining 10x annually. Agents will go free. The question isn't whether agents will be ad-supported - it's who builds the ad network.

The conversational AI market is $18B today, projected to reach $32-44B by 2030. OpenAI's ChatGPT ads hit $100M ARR in 6 weeks, validating the category. ZeroClick raised $55M. Kontext raised $10M. The market is real and capital is flowing.

Operon is building the open, quality-weighted ad network for the agent ecosystem. The same way AdSense served the long tail of web publishers that Yahoo couldn't reach, and AdMob served the long tail of mobile developers that carrier decks ignored.