Self-service has always been at the center of Platform Engineering. Over the years, we’ve built mature ecosystems around reusable infrastructure — Helm chart repositories, Terraform module registries, internal developer portals, and API catalogs — all designed to help teams move faster without reinventing the wheel.

But as AI agents become active participants in engineering workflows, a new question starts to emerge:

Why don’t we have a marketplace for agent capabilities?

Today, most AI agents still operate with tightly coupled tool configurations, hard-coded integrations, and manually curated access lists. Every new capability often requires custom onboarding, duplicated integration logic, and platform-level coordination. As the number of agents and tools grows, this quickly becomes difficult to scale.

This talk introduces the concept of an Agent Capability Provider (ACP) Marketplace — a self-service ecosystem where platform teams can publish reusable capabilities that agents can discover and consume dynamically through standardized protocols.

At the center of this model is the ACP Protocol, which enables a plug-and-play architecture for agent tooling. Instead of every team building custom integrations, capabilities can be registered once and exposed through a governed marketplace layer connected via the MCP Gateway. This creates a standardized contract for how tools are described, discovered, authenticated, and invoked across the organization.

We’ll explore what it means to design a true capability marketplace for autonomous systems — where agents can evaluate and select tools based on factors such as reliability, latency, cost, permission scope, and operational health.

The session will also demonstrate how dynamic capability discovery changes the way agents operate. Rather than relying on static tool lists embedded into prompts or orchestration code, agents can query the marketplace in real time to understand what capabilities are available and determine the best option for a given task.

Beyond architecture, we’ll discuss the operational and business impact of this model. One of the biggest hidden problems in enterprise AI systems today is what I call “Discovery Debt” — teams repeatedly rebuilding integrations because they have no centralized understanding of what capabilities already exist.

By treating agent tools as products within a governed marketplace, platform teams gain visibility into usage patterns, effectiveness, operational cost, and adoption across the AI fleet. This creates measurable ROI for platform engineering while also helping retire unused or low-value capabilities over time.

Key takeaways include:

• Understanding the ACP Protocol and plug-and-play capability architecture
• Designing a governed marketplace for reusable agent capabilities
• Enabling dynamic capability discovery through MCP Gateway integration
• Reducing “Discovery Debt” across AI and platform teams
• Measuring platform ROI through capability usage, effectiveness, and lifecycle management
• Building scalable self-service ecosystems for autonomous engineering systems