Adform Debuts 29 Agentic Skills: A New Era of Read-Only AI Oversight for Programmatic Advertising

In a significant development for programmatic infrastructure, Adform’s Solutions Engineering team has released a suite of 29 "agentic skills" designed to bridge the gap between AI-driven natural language processing and the complex data architecture of the Adform FLOW demand-side platform (DSP). Published via a public GitHub repository under the name agentic-skills, this release provides a structured framework for AI systems to query, audit, and analyze campaign data without the risks typically associated with autonomous execution.

By leveraging the Model Context Protocol (MCP)—an open standard initially developed by Anthropic—these skills allow users to interact with Adform’s data through conversational interfaces. Crucially, Adform has adopted a strictly "read-only" posture, ensuring that while AI can provide deep diagnostic insights, it lacks the technical capability to alter live campaign parameters, budgets, or creative assets.

The Architecture of AI-Driven Insight

At the heart of the release are 29 distinct markdown files, each serving as a blueprint for an AI agent. According to the repository’s documentation, these files do not contain executable binary code; rather, they provide the "purpose, illustrative GraphQL queries, usage constraints, and presentation guidance" required for an AI to interpret and act upon a user’s query.

How the Skills Operate

When an AI agent is connected to the Adform GraphQL MCP server, it parses these skill files to understand which data points are required to answer a user’s prompt. For example, if a media buyer asks, "Which campaigns are failing our naming convention?", the AI identifies the relevant "Taxonomy Governance" skill file, executes the necessary GraphQL query, and generates a structured response.

This design is deliberately granular. The taxonomy governance skill, for instance, uses regex patterns (such as ^[A-Za-z]+_d4_Q[1-4]_[A-Za-z]+$) to validate campaign nomenclature. If a campaign deviates from the required format—Brand_Year_Quarter_Objective—the AI flags the error. It also checks for missing labels (e.g., Market, Product, or Funnel Stage) and generates a compliance report that prioritizes items needing immediate attention.

The instructions for the AI are precise, dictating not just the logic, but the presentation: the output must be formatted as compliance tables, leading with the count of non-compliant items. This "report-first" approach is mirrored across all 29 skills, which cover domains ranging from budget risk monitoring and inventory forecasting to frequency cap auditing and channel conflict detection.

Chronology of Development and Adoption

The release of these skills is the latest milestone in a two-year push by Adform to modernize its API interaction layer.

• Mid-2025: Adform begins intensive work on the MCP Gateway, a foundational layer designed to facilitate secure, standardized communication between the Adform platform and external AI agents.
• Late 2025 – Early 2026: The industry experiences a surge in MCP adoption. Google releases its Ads API MCP server in October 2025, while competitors like Microsoft and Yahoo DSP move quickly to roll out their own agentic frameworks.
• Early 2026: Adform’s internal engineering team, led by Suren Silva, Marcel Ehrlitzer, and Hans Jirschik, begins iterative commits to the agentic-skills repository.
• March 2027: The official public release of the 29-skill collection is announced. The GitHub repository, which now boasts 26 commits, indicates that the project remains under active refinement, with recent updates focused on synchronizing audience discovery skills across distribution channels.

Supporting Data and Technical Distribution

The repository is engineered for versatility, supporting two primary deployment pathways to ensure compatibility with various AI ecosystems.

Deployment Paths

1. dist/claude/: This directory contains a packaged plugin optimized specifically for Anthropic’s Claude Code, a command-line interface tool. Users can install these via simple commands like claude plugin marketplace add adform/agentic-skills.
2. dist/generic/: Designed for broader compatibility, these files follow the agentskills.io standard, allowing developers to integrate Adform’s skills into alternative agentic frameworks or custom-built AI environments.

Rate Limiting and Governance

The technical documentation includes built-in safeguards, such as a requirement for GraphQL calls to be spaced one to two seconds apart. This prevents API abuse and ensures that the agentic layer remains performant even when handling large-scale data queries across complex account structures.

Official Perspectives and Industry Context

Suren Silva, VP of Global Solutions Engineering at Adform, emphasized that this release is just the beginning. In a post on LinkedIn, Silva highlighted the work of his team—specifically Marcel Ehrlitzer, Hans Jirschik, and Maja Sokołowska—and explicitly stated, "A LOT more skills to come."

The "Read-Only" Debate

Adform’s decision to restrict these skills to read-only access places the company firmly in the "conservative" camp of AI integration. This aligns with the strategic choices made by Google, Amazon Ads, and Microsoft, all of whom have introduced read-only MCP servers for their respective advertising platforms.

This stands in contrast to approaches like that of Meta, which launched its Ads AI Connectors with "write" capabilities, allowing agents to manipulate campaigns. The Adform approach is seen by many industry analysts as a safer, "human-in-the-loop" model. By keeping the AI in an advisory role—where it flags discrepancies but requires a human trafficker to implement the fix—Adform effectively sidesteps the risks of "hallucinated" budget spends or catastrophic targeting errors that have plagued less-governed autonomous systems.

Implications for Marketers and Agencies

For the typical media buyer, the impact of this release is not a total overhaul of the daily workflow, but rather a significant reduction in operational friction.

From Manual Auditing to Instant Insight

Historically, identifying naming convention drift or missing labels required hours of manual labor—pulling reports, exporting them to spreadsheets, and applying formulas to verify compliance. By delegating these diagnostic tasks to an AI, teams can now shift their focus from finding problems to solving them.

Governance and Risk Mitigation

The "read-only" boundary provides a necessary layer of governance. As noted by industry experts like Ari Paparo, the proliferation of agentic protocols brings with it significant risk. By limiting agents to reporting and flagging, Adform creates a "sandbox" where AI can demonstrate value without the risk of overspending or misconfiguring campaigns.

However, the human element remains paramount. The skills, particularly those involved in naming convention correction, offer "suggestions." If an AI incorrectly interprets a complex or edge-case campaign name, the human trafficker retains the final say. The release does not replace the operator; it empowers them with a more efficient, real-time diagnostic tool.

Future-Proofing the Stack

The release also underscores the growing importance of the Model Context Protocol as a universal language for ad tech. With firms like PubMatic, DoubleVerify, and Yahoo already heavily invested in MCP, the industry is moving toward a standardized ecosystem where agents can perform cross-platform audits. While Adform’s 29 skills are currently specific to their own DSP, the adoption of an open protocol suggests that in the future, these skills could eventually be part of a broader, interoperable suite of tools.

Conclusion

Adform’s release of 29 agentic skills marks a pivotal step in the evolution of programmatic advertising. By prioritizing transparency, read-only governance, and structured data interaction, the company has provided a model for how complex ad tech platforms can safely adopt AI.

As the repository continues to be updated and more skills are added, the value proposition will likely shift from simple diagnostic reporting to more complex, multi-layered predictive analysis. For now, the focus is on utility and safety—giving marketers the clarity they need to manage their campaigns with greater precision, while keeping the "keys to the kingdom" firmly in the hands of human operators. Whether these skills will lead to widespread adoption depends on the reliability of the outputs and the ability of Adform to maintain its momentum in an increasingly crowded and competitive agentic landscape.