On June 24, 2026, a media agency buyer in Prague typed a campaign brief into ChatGPT - and the rest happened without a human touch. Inventory was matched, terms were negotiated, and the purchase was executed end-to-end by autonomous AI agents. No spreadsheets. No procurement calls. The Czech Republic's first fully agentic ad buy was complete.

What happened and who was involved

R2B2, an advertising technology company headquartered in Prague and founded in 2014, facilitated the transaction between Omnicom Media, the Czech operation of the global Omnicom Group, and MAFRA, one of the country's leading digital publishers. The advertiser behind the campaign was Skylink, a Czech satellite pay-TV service that operates as part of the European television group Canal+.

According to R2B2, this marks the first time any autonomous AI agent has executed a complete ad space purchase in the Czech Republic without manual intervention at the transactional stage. The pilot is the first public demonstration of what R2B2 calls "agentic advertising" in the CEE region.

The scale of the pilot is modest - one campaign, one buyer, one publisher. What matters is the architecture it demonstrates, and what that architecture implies for how media trading could operate at scale across European markets.

How the transaction actually worked

The mechanics deserve close attention. Rather than navigating inventory spreadsheets or negotiating terms via email or phone, the buyer at Omnicom Media used ChatGPT as a conversational interface. By describing the campaign's objectives and targeting requirements in natural language, the buyer triggered a chain of automated actions that culminated in a completed media purchase.

R2B2 played the role of what it describes as the "sell-side agent" - the technical intermediary between the buying AI and MAFRA's existing ad server. The connection between these two sides was made possible through a proprietary, internally developed MCP (Model Context Protocol) gateway that R2B2 built on top of its monetization infrastructure.

MCP, or Model Context Protocol, was originally developed by Anthropic and subsequently donated to the Linux Foundation. It has emerged over the past 18 months as a connective standard across the programmatic advertising industry. PPC Land has tracked its adoption across advertising infrastructure throughout 2025 and into 2026, with platforms including AdRoll, PubMatic, FreeWheel, Yahoo DSP, and DoubleVerify all building MCP-based integrations for campaign management, deal diagnostics, and optimization workflows.

In the R2B2 implementation, the buying agent - operating through ChatGPT - communicated with R2B2's MCP gateway using a separate protocol layer called the Ad Context Protocol (AdCP). The AdCP is an open-source specification built on top of MCP, launched on October 15, 2025 by a coalition of six founding companies - PubMatic, Scope3, Swivel, Triton Digital, Optable, and Yahoo - specifically to standardize how AI agents communicate across advertising platforms. According to R2B2, its pilot used AdCP to govern the interaction model between the buying agent and its sell-side gateway.

The entire solution runs as a managed cloud-based SaaS service, which R2B2 says effectively removes technical barriers for publishers that want to participate in agentic transactions. Crucially, the integration is entirely server-to-server. No client-side scripts are involved. According to R2B2, this means zero latency impact on MAFRA's web pages and no disruption to the user experience of the publisher's audience.

This server-to-server architecture is a meaningful technical detail. Client-side integrations carry real costs: additional JavaScript on a page adds load time, and publishers with strict performance targets have historically resisted adding new tags. A server-to-server approach sidesteps that objection entirely.

The role of ChatGPT as the buying interface

The decision to use ChatGPT as the buyer-facing interface is worth examining separately from the infrastructure layer. ChatGPT is not a specialist advertising tool. It is a general-purpose large language model assistant. By connecting it to R2B2's MCP gateway through the AdCP, Omnicom Media demonstrated that the buyer's interface can be any AI assistant capable of using MCP - not a bespoke advertising platform.

This matters because it implies a future in which the buying side of programmatic advertising becomes accessible through whatever conversational AI tool a media planner already uses. The specialist interface - the DSP dashboard, the proposal form, the spreadsheet request for proposal - becomes optional if the MCP gateway is in place on the sell side and the buyer's AI assistant can discover and call it.

The agentic advertising trajectory has been visible across the industry since mid-2025, when analysts began modelling scenarios in which AI agents bypass traditional demand-side platforms entirely. In this Czech pilot, there is no DSP in the transaction chain at all. The buying agent communicated directly with R2B2's sell-side infrastructure.

Human oversight built into the workflow

The pilot did not remove humans from the process entirely. According to R2B2, publishers retain the final say: every campaign and creative asset must be manually approved by MAFRA before it goes live. This is what the industry describes as a "human-in-the-loop" design - high automation in the discovery and negotiation stages, human approval at the point of activation.

"Our goal was to create a secure, intuitive environment where both parties can operate seamlessly without disrupting established approval workflows," said Lukáš Alexandr, CTO at R2B2.

That framing reflects a structural choice about where in the workflow human oversight adds most value. In the pilot, humans are not needed to navigate inventory or negotiate terms - but they retain control over what runs, and where. MAFRA's David Korn, Commercial Director, confirmed that the pilot successfully verified their infrastructure's readiness for AI agents, and that automated transactions posed no risk to web operations, according to R2B2.

The human-in-the-loop design has become something of a consensus pattern among early agentic advertising implementations. Yahoo DSP integrated agentic AI capabilities on January 6, 2026, with its troubleshooting agent able to diagnose and recommend corrective actions but requiring human approval to execute them. That sequencing - agent proposes, human approves - is the same logic R2B2 applied at the point of campaign activation.

Jindrich Jiracek, Head of Performance at Omnicom Media, described the approach this way in remarks attributed to him by R2B2: "We view agentic buying as the natural evolution of digital advertising. This technology has the potential to dismantle long-standing barriers between campaign briefing, inventory selection, and execution. It opens new frontiers for automation and collaboration while preserving the human strategic oversight that is vital for our decision-making."

The comment reflects a specific operational problem that the agentic approach is intended to address. In conventional media buying, the gap between a campaign brief and live inventory is often measured in days - multiple rounds of communication, internal approval steps, counter-proposals from publishers, creative trafficking. Agentic tools compress that timeline.

Tomas Jelinek, Acquisition Marketing Manager at Skylink, offered a direct-advertiser perspective in remarks attributed to him by R2B2: "At Skylink, we are continuously looking for new ways to reach relevant audiences more effectively while maintaining a high-quality media environment. We see the agentic buying pilot as an interesting step toward greater automation, faster work with available inventory, and more precise alignment of the campaign with the appropriate context. What is crucial for us is that even when using advanced technology, we maintain control over where and how our brand is displayed."

The emphasis on brand control is consistent with how advertisers across the market have positioned themselves relative to agentic tools. Autonomy in execution is accepted; autonomy in brand placement decisions is not.

The AdCP standard and where this fits in the broader landscape

The use of AdCP in the Czech pilot is notable in itself. The Ad Context Protocol launched on October 15, 2025, with considerable industry debate about whether it addressed fundamental structural problems or merely added another protocol layer on top of existing fragmentation. Prominent critics at the time noted that major demand-side platforms - Google, Amazon, The Trade Desk, and Microsoft - did not sign up for AdCP, raising questions about how widely the standard could realistically be adopted.

Since then, evidence of AdCP-based live activity has been accumulating. VIOOH's Seller Agent automated more than 100 curated digital out-of-home private marketplace deal packages in the first half of 2026 using infrastructure aligned with AdCP. Broadsign and Draft Digital ran the first reported fully agentic out-of-home campaign in May 2026 using AdCP. PubMatic's AgenticOS launched in January 2026 with live campaigns running across WPP Media, Butler/Till, and MiQ.

The R2B2 pilot adds a new type of deployment to that record. Where previous AdCP-based live activity has primarily occurred in the United States and in digital out-of-home inventory, the Czech campaign demonstrates the protocol operating for standard digital display advertising in a Central European market, with a mid-size regional publisher and a pay-TV advertiser - neither of which fits the early-adopter profile of the infrastructure companies that designed the protocol.

R2B2 is not itself an AdCP founding member or a large platform. It is a technology company operating across 46 markets with a concentration in CEE, which it built through proprietary monetization tools since its founding in 2014. Its MCP gateway is an internally developed product rather than a direct implementation of a major platform's infrastructure. That independence is part of what makes the Czech pilot meaningful. If agentic advertising requires either deep pockets or founding-member status in a major protocol consortium, its adoption will remain slow. If a mid-size regional ad tech company can build a working implementation that executes a live transaction, the approach is more accessible than the infrastructure debate suggests.

R2B2's scale and what comes next

R2B2 describes itself as operating infrastructure across 46 markets. The company's stated focus is the CEE region, where it has developed relationships with publishers and agencies over more than a decade. That regional footprint is relevant to what happens after a pilot.

According to R2B2, the plan is to scale the platform by onboarding additional publishers and continuing to refine the integration. The company has not disclosed timelines or the specific publishers it is targeting. Lukáš Alexandr, CTO at R2B2, described the direction of travel in remarks attributed to him: "Artificial intelligence and autonomous, agentic purchasing are the clear future of digital advertising. Publishers and advertisers alike need to prepare, as this method of trading ad space is poised to become the new market standard."

That forecast is consistent with the pattern that PPC Land has documented across the industry through the first half of 2026. Agentic advertising is no longer theoretical. It is running live campaigns. What remains genuinely uncertain is the pace at which the operational model spreads from early participants - technically sophisticated agencies, platform-native publishers, and dedicated ad tech companies building the infrastructure - to the broader market of buyers and sellers who have not yet begun preparing.

The IAB Europe CTV programmatic guide from April 2026, which included a Czech case study involving Skylink, OMC Czech Republic, and Adform, documented a market already engaging with complex programmatic structures. That the same Czech market now has the first agentic direct buy on record is not a coincidence - it reflects genuine activity among Czech market participants in programmatic infrastructure, across both display and connected television.

Omnicom itself has been moving on multiple agentic fronts simultaneously. At Cannes Lions 2026 in late June, Adobe announced that Omnicom is deploying an AI Agentic Operating Model across four industry verticals - automotive, pharmaceuticals, retail, and financial services - as part of a broader enterprise implementation of Adobe's agentic infrastructure. The Czech pilot and the Cannes announcement are separate tracks, but they reflect the same strategic direction inside the holding company.

Why this matters for European markets

European markets have followed a different trajectory from the US in programmatic advertising, shaped by GDPR, market fragmentation across languages and regulatory environments, and a publisher landscape that differs substantially from the open-web display market in North America.

The Czech pilot is the first publicly documented instance of fully autonomous agentic ad buying in a Central European market. That is a specific geographic and regulatory context. MAFRA operates in an environment where data handling, consent, and brand safety standards follow EU frameworks. The R2B2 integration is server-to-server, meaning it does not involve client-side data collection. Combined with the human approval requirement for creative assets, the architecture as described is designed to operate within publisher-controlled environments rather than requiring publishers to extend trust to opaque automated systems.

Whether the model spreads across the CEE region depends partly on how other publishers assess their own infrastructure readiness. David Korn's confirmation that the pilot verified MAFRA's infrastructure readiness for AI agents is a data point other publishers in the region will be watching.

Timeline

Summary

Who: R2B2 (advertising technology company, Prague, founded 2014, operating across 46 markets), Omnicom Media (Czech operation of the global Omnicom Group), MAFRA (Czech digital publisher), and Skylink (Czech satellite pay-TV service, part of Canal+).

What: The Czech Republic's first fully autonomous AI agent ad buy, executed through a ChatGPT interface connected to R2B2's proprietary MCP gateway and the Ad Context Protocol, with MAFRA as publisher and Skylink as the end advertiser. The transaction covered inventory identification, negotiation, and execution without manual intervention in the transactional stages - though human approval of creative assets was retained at MAFRA's end.

When: The transaction was executed on June 24, 2026, and publicly announced the same day.

Where: Prague, Czech Republic. The technical transaction was server-to-server, involving no client-side scripts and causing no latency impact on MAFRA's web properties.

Why: R2B2 designed the pilot to demonstrate that agentic advertising - autonomous AI-driven media buying using open protocols - is technically feasible for standard digital advertising in a Central European market, not only in the US display and CTV environments where most agentic deployments have occurred so far. For Omnicom Media and Skylink, the pilot tests whether AI-mediated execution can compress the timeline between campaign brief and live inventory without sacrificing brand control or publisher approval workflows.