Agentic APIs
How AI agents are built, connected, and orchestrated - across SDKs, protocols, CLIs, and platforms.
Overview
The agentic AI landscape in 2025-2026 has three layers:
- Protocols - How agents talk to tools and each other (MCP, A2A, ACP)
- SDKs - How you build agents in code (Claude Agent SDK, OpenAI Agents SDK, Copilot SDK)
- Standards - Skills as an open standard across IDEs (Dec 2025), portable agent workflows
- Products - Agentic tools you use directly (Claude Code, Copilot CLI, Cursor, Aider)
This module covers all three layers, with emphasis on the patterns that transfer across ecosystems.
Contents
- Concepts - The mental models you need
- The Agentic Loop - The core pattern every agent uses
- Protocols: MCP vs A2A vs ACP - How agents connect to tools and each other
- SDK Comparison - Claude, OpenAI, Copilot SDKs side by side
- Quickstart - Get hands-on fast
- Build an Agent with Claude - Agent SDK + tool use in 50 lines
- Build an Agent with OpenAI - Agents SDK with handoffs
- Build with Copilot SDK - Copilot SDK multi-language quickstart
- Deep Dives - Advanced patterns
- Multi-Agent Orchestration - Handoffs, delegation, parallel agents
- Agent Teams - Multiple agents coordinating on shared tasks
- Skills as Open Standard - Portable agent workflows across IDEs
- Agentic CLIs - Claude Code, Copilot CLI, Aider architecture
- GitHub Agentic Workflows - AI in CI/CD
- Notes - Gotchas, opinions, and personal reference