Architecting Developer Platforms for AI Agents: Roles and Requirements
This article explores three distinct roles AI agents can play within a developer platform: as consumers, internal components, and manageable resources. It outlines the architectural requirements for platforms to support each role, emphasizing the need for robust APIs, governed context layers, orchestration, and self-service capabilities. The discussion provides insights into integrating AI agents into existing engineering workflows and platform designs.
Read original on The New StackThe integration of AI agents into developer platforms is rapidly evolving, aiming to boost engineering productivity. This shift necessitates careful architectural considerations to effectively support these agents. The article identifies three primary roles AI agents can assume, each with specific implications for platform design.
Role 1: AI Agents as Platform Consumers
In this model, an AI agent acts much like a human user, consuming platform services to accomplish its tasks. This includes reading system context (e.g., service catalogs, ownership, dependencies, standards) and executing self-service actions (e.g., spinning up preview environments, running tests).
Platform Requirements for Consumer Agents
To enable this role, the developer platform must provide: - An API- and MCP-first interface for programmatic access. - A governed context layer (e.g., a "context lake") acting as a single source of truth for all system information. - A suite of self-service actions that agents can invoke.
Role 2: AI Agents as Internal Platform Components
Here, AI agents are embedded within platform workflows, triggered by events rather than direct human requests. They act as a non-deterministic step alongside traditional deterministic steps in an orchestration engine. An example is an agent automatically creating pull requests to remediate vulnerable dependencies identified by a nightly scan.
Platform Requirements for Internal Component Agents
Supporting agents as internal components requires: - An orchestration layer capable of running and managing agents within workflows. - An agent registry for discovery and retrieval of specific agents. - Unique identities and credentials for each agent to ensure proper logging and security. - Human-in-the-loop steps for critical or high-risk actions to maintain oversight.
Role 3: AI Agents as Manageable Resources (AgenticOps)
This role treats AI agents, their underlying LLMs, and skills as first-class resources, akin to services, databases, or environments. The platform is responsible for provisioning, governing, and managing their entire lifecycle, often through a "golden path" self-service experience. Engineers can request a specific agent (e.g., an on-call triage agent) and have it provisioned with the right model, tools, and context.
Platform Requirements for AgenticOps
For AgenticOps, the platform needs: - A self-service provisioning path to deploy agent runtimes. - Mechanisms to issue identity and scoped credentials. - Functionality to wire in approved context. - An agent registry to publish and discover agents for reuse.