Develop Autonomous AI Agents and Systems

This is a router workflow for building AI agents - single autonomous agents, multi-agent systems, orchestration, tool integration, and memory - structured as seven sequential phases, each pointing to specific sub-skills and copy-paste prompts to invoke them. Phase 1, Agent Design, uses ai-agents-architect and autonomous-agents to define purpose, capabilities, tool integration plans, memory design, and success metrics. Phase 2, Single Agent Implementation, uses autonomous-agent-patterns and autonomous-agents to choose a framework, implement agent logic, wire in tools and memory, and test behavior. Phase 3, Multi-Agent System, uses crewai and multi-agent-patterns to define agent roles, set up inter-agent communication, configure orchestration, and implement task delegation. Phase 4, Agent Orchestration, uses langgraph and workflow-orchestration-patterns to design a stateful workflow graph with conditional branches and persistence. Phase 5, Tool Integration, uses agent-tool-builder and tool-design to identify tool needs, design interfaces, implement tools, and add error handling. Phase 6, Memory Systems, uses agent-memory-systems and conversation-memory to design short-term, long-term, and entity memory with retrieval testing. Phase 7, Evaluation, uses agent-evaluation and evaluation to define criteria, build test scenarios, measure performance, and test edge cases. The overall architecture flows User Input -> Planner -> Agent -> Tools -> Memory -> Response, with the planner decomposing the request, the LLM core reasoning over it, tools executing actions, and short/long-term memory feeding back into the loop.

Use this when building an autonomous AI agent, a multi-agent system, agent orchestration, tool integration for agents, or an agent memory system - as the entry point that routes to the specific sub-skill for whichever phase the work is currently in, rather than as a standalone implementation guide itself.

A quality-gate checklist should be satisfied before considering the work done: agent logic working, tools integrated, memory functional, orchestration tested, and evaluation passing. Related workflow bundles cover adjacent scope: ai-ml for general AI/ML development, rag-implementation for RAG systems specifically, and workflow-automation for broader workflow patterns.

Input is the current phase of an agent-building effort (design, single-agent implementation, multi-agent system, orchestration, tools, memory, or evaluation). Output is a copy-paste prompt invoking the correct named sub-skill for that phase (e.g. Use @langgraph to create stateful agent workflows), plus the specific action checklist for that phase.

Routes to ai-agents-architect, autonomous-agents, autonomous-agent-patterns, crewai, multi-agent-patterns, langgraph, workflow-orchestration-patterns, agent-tool-builder, tool-design, agent-memory-systems, conversation-memory, and agent-evaluation as its constituent sub-skills, and links to ai-ml, rag-implementation, and workflow-automation as related workflow bundles.

Developers building AI agents who need a phase-by-phase entry point routing to the right specialized skill for design, implementation, orchestration, tooling, memory, or evaluation.

User Input -> Planner -> Agent -> Tools -> Memory -> Response
              |          |        |        |
         Decompose   LLM Core  Actions  Short/Long-term