Agentic AI Standards and Engineering ROI
Manik Sertani discusses the Agentic AI Foundation's impact on MCP and A2A adoption. Learn how engineering leaders can manage AI-generated code bottlenecks, shift from code writing to factory oversight, and measure true ROI in agentic workflows.
The Rise of Neutral Agentic Standards
The Agentic AI Foundation (AAIF) is reshaping the AI infrastructure landscape by consolidating key protocols like MCP, A2A, and Goose under neutral governance. This strategic shift has eliminated vendor-specific trust barriers, leading to a five-fold increase in MCP SDK downloads and accelerating enterprise adoption. The integration of A2A, originally from Google, further validates the ecosystem's vendor neutrality, allowing diverse agent protocols to coexist and interoperate without proprietary lock-in. For engineering leaders, this standardization reduces integration risk and fosters a collaborative open-source environment where innovation is driven by community consensus rather than corporate dominance.
The New Engineering Bottleneck: Review and Rework
Despite AI accelerating code generation, the software development lifecycle faces a critical bottleneck in code review. Data indicates that up to 70% of agentic pull requests in mature organizations remain unmerged, creating a "slop cannon" effect that overwhelms human reviewers. The cost of AI-generated code has shifted downstream, making traditional review processes inefficient. Organizations must redefine quality assurance by enforcing author accountability for AI-generated code or adopting issue-based contribution models where humans provide strategic direction and agents execute the implementation. This shift requires a cultural change where engineers focus on verifying outcomes rather than reviewing line-by-line code.
Measuring ROI in Agentic Workflows
A significant challenge for engineering leaders is quantifying the return on investment of agentic tools. Without visibility into token spend and task-level costs, it is difficult to determine which workflows deliver value. Implementing centralized gateways, such as Agent Gateway, allows teams to track inference costs, pin metadata to tasks, and identify inefficiencies. This data-driven approach enables leaders to optimize model selection, reduce waste, and prove the durable value of AI initiatives. As the "software factory" model emerges, engineers must transition from individual contributors to factory foremen, overseeing automated pipelines and ensuring that AI actions align with business objectives. The future of engineering lies in system-level thinking, where the focus shifts from writing code to designing, verifying, and governing autonomous systems.
Key insights
-
Neutral governance is the primary driver of adoption for AI protocols. Projects like MCP and A2A see exponential growth when moved to foundations like AAIF, as it removes trust barriers for competitors and distributors.
Impact: Enterprises are more likely to adopt standardized, vendor-neutral protocols, reducing integration costs and long-term technical debt.
-
The value of open source contribution has shifted from code submission to idea sharing. As AI makes code generation cheap, maintainers prioritize high-quality issues and strategic insights over raw pull requests.
Impact: Contributors must adapt by focusing on problem definition and solution design, while maintainers curate communities around strategic alignment rather than code volume.
-
AI-generated code creates a review bottleneck, with high unmerged rates for agentic PRs. Traditional code review processes are insufficient for handling the volume and nature of AI output.
Impact: Teams risk increased rework costs and delayed delivery if they do not implement new verification methods or enforce author accountability for AI-generated code.
-
Visibility into token spend and task-level costs is critical for measuring AI ROI. Tools like Agent Gateway enable teams to track inference costs and identify inefficiencies in agentic workflows.
Impact: Data-driven cost tracking allows leaders to optimize model usage, justify AI investments, and eliminate low-value workflows.
-
The role of the engineer is evolving from code writer to factory foreman. Engineers must focus on system design, verification, and strategic oversight of automated pipelines rather than manual coding.
Impact: Organizations must invest in upskilling engineers for system-level thinking to fully leverage the benefits of the software factory model.
Action items
-
Adopt neutral, open-source standards like MCP and A2A for agent communication to ensure vendor independence and long-term interoperability.
Impact: Reduces vendor lock-in risk and accelerates adoption of best-in-class agent infrastructure.
-
Implement centralized AI gateways to track token spend and associate costs with specific tasks or workflows.
Impact: Provides the data necessary to calculate accurate ROI and identify cost-saving opportunities in agentic operations.
-
Redefine code review policies for AI-generated code by enforcing author accountability or shifting to issue-based contribution models.
Impact: Mitigates the review bottleneck and ensures that AI-generated code meets quality standards before merging.
-
Train engineers to adopt a factory foreman mindset, focusing on system design, verification, and strategic oversight of automated processes.
Impact: Aligns workforce skills with the evolving software factory model, maximizing the value of AI automation.
-
Maintain model independence by using routing layers that allow easy switching between LLMs based on cost and performance metrics.
Impact: Ensures flexibility and cost optimization as the AI model landscape evolves rapidly.
Quotes
“We have over a billion downloads right now of the MCP SDKs.”
“The value has moved. Value is in sharing ideas, in sharing experiences, in sharing thoughts.”
“I will skip the code review part. I'll have an AI do the code review, provided you as the author read your own code that you've generated and approve it.”