Agents¶
Concepts for building AI agents and implementing tool use.
Agents by Platform¶
Each platform has its own agent system. These guides show how to build agents on each one — from no-code options for business users to full developer frameworks.
| Platform | Approach | Description | Guide |
|---|---|---|---|
| Claude | Subagents | Build and activate subagents via natural language | Building Agents on Claude |
| Claude | Agent Teams | Code-first multi-agent coordination | ↑ |
| Claude | Claude in Chrome | Chrome browser orchestration (macOS only) | ↑ |
| Claude | Agents SDK | Code-first approach in Python with SDK | ↑ |
| OpenAI | ChatGPT Agent Mode | Single-agent orchestration in ChatGPT | Building Agents on OpenAI |
| OpenAI | Atlas Browser | Browser orchestration (macOS only) | ↑ |
| OpenAI | AgentKit | Visual builder for conversational agents | ↑ |
| OpenAI | Agents SDK | Code-first approach with Python | ↑ |
| OpenAI | Frontier | Enterprises orchestrating custom, OpenAI, and third-party agents | ↑ |
| Workspace Studio | Simple agent flows for Workspace apps | Building Agents on Google | |
| Gemini Enterprise Agent Designer | Build agents and workflows visually — enterprise offering | ↑ | |
| Gemini in Chrome | AI-powered browser workflows | ↑ | |
| Vertex AI & ADK | Code-first approach in Python | ↑ | |
| M365 Copilot | Copilot & Copilot Studio | Build agents visually and deploy to M365 surfaces | Building Agents on M365 Copilot |
For guidance on deciding whether your workflow needs an agent, see Build Workflows > Agents.
Topics¶
- SDK Building Block — Frameworks and toolkits for building agents and AI workflows in code (Claude Agent SDK, OpenAI Agents SDK, Google ADK, LangGraph, and more)
- Agent Capability Patterns — Seven architectural patterns that make agents effective: Reflection, Tool Use, Planning, Multi-Agent Collaboration, Memory, Guardrails, and Human-in-the-Loop
- Agent Orchestration Patterns — Eight coordination topologies for multi-agent systems: Sequential, Parallel, Router, Hierarchical, Handoff, Evaluator-Optimizer, Group Chat, and Decentralized
Key Concepts¶
- Function Calling - Enabling models to call external functions
- Tool Use - Defining and executing tools
- Agent Loops - Iterative reasoning and action
- ReAct Pattern - Reasoning and acting
- State Management - Tracking conversation and tool state
- Error Handling - Graceful failure and recovery
Related¶
- Agentic Building Blocks
- SDK — frameworks for building agents in code
- API — programmatic interfaces agents call under the hood
- AI Use Cases — what teams build with agents, organized by six primitives
- Automation Use Cases — the highest-autonomy use cases powered by agents
- Workflow Architecture Patterns — seven patterns from augmented LLMs to autonomous agents
- Patterns