Develop Autonomous AI Agents and Systems

This is a granular workflow bundle for building AI agent systems end to end, structured as seven sequential phases. Each phase names the specific sub-skills to invoke, an action checklist, and a copy-paste prompt to trigger it.

Phase 1, Agent Design, calls ai-agents-architect and autonomous-agents to define the agent's purpose, capabilities, tool integration plan, memory system, and success metrics. Phase 2, Single Agent Implementation, uses autonomous-agent-patterns and autonomous-agents to choose a framework, implement the agent's logic, add tool integration, configure memory, and test behavior. Phase 3, Multi-Agent System, invokes crewai and multi-agent-patterns to define agent roles, set up agent-to-agent communication, configure orchestration, implement task delegation, and test coordination. Phase 4, Agent Orchestration, uses langgraph and workflow-orchestration-patterns to design a workflow graph, implement state management, add conditional branches, configure persistence, and test the resulting workflows. Phase 5, Tool Integration, calls agent-tool-builder and tool-design to identify tool needs, design tool interfaces, implement the tools, and add error handling. Phase 6, Memory Systems, uses agent-memory-systems and conversation-memory to design the memory structure and implement short-term, long-term, and entity memory. Phase 7, Evaluation, invokes agent-evaluation and evaluation to define evaluation criteria, build test scenarios, measure performance, and test edge cases.

The workflow models the overall agent architecture as a pipeline: user input flows through a planner that decomposes the task, into the agent's LLM core, out to tools for actions, and through short- and long-term memory before producing a response.

Use this workflow when building autonomous AI agents, creating multi-agent systems, implementing agent orchestration, adding tool integration to an agent, or setting up an agent memory system. It is meant as a sequencing guide across the full agent-development lifecycle, not a single-purpose tool - apply it when the work spans multiple phases (design through evaluation), not for an isolated one-off task like a single bug fix.

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

Input is the user's task description; the Planner block decomposes it, the Agent's LLM Core reasons over it, Tools execute actions, and Short/Long-term memory feeds back into the response. Each of the seven phases closes with its own quality-gate checklist before moving to the next: agent logic working, tools integrated, memory functional, orchestration tested, and evaluation passing.

The bundle is designed to compose with related workflow bundles: ai-ml for broader AI/ML development work, rag-implementation for RAG systems, and workflow-automation for general workflow automation patterns.

Developers and teams building autonomous agents or multi-agent systems who need a structured, phase-by-phase path from architecture design through single-agent implementation, multi-agent coordination, orchestration, tool integration, memory, and evaluation, rather than starting from a blank page and improvising the sequence.