Skip to content

Adjacent Agent Ecosystem Map

This guide focuses on AI engineering workflows, harnesses, and app frameworks. The broader ecosystem also includes agent SDKs, managed agent services, low-code automation, data frameworks, and observability platforms.

The goal of this page is to prevent category confusion.

Ecosystem by layer

mermaid
flowchart TB
    A[AI engineering ecosystem] --> B[Delivery workflow]
    A --> C[Coding agent harness]
    A --> D[Agent app framework]
    A --> E[Managed agent platform]
    A --> F[Low-code workflow automation]
    A --> G[Data/RAG framework]
    A --> H[Evals and observability]

    B --> B1[Spec Kit / OpenSpec / AI-DLC / GSD / Superpowers]
    C --> C1[Codex CLI / Claude Code / Hermes]
    D --> D1[LangChain / LangGraph / AutoGen / CrewAI / Google ADK / OpenAI Agents SDK]
    E --> E1[Azure AI Foundry Agents / Amazon Bedrock Agents]
    F --> F1[Dify / n8n]
    G --> G1[LlamaIndex / Haystack / Semantic Kernel connectors]
    H --> H1[LangSmith / Langfuse / Phoenix / OpenTelemetry]

Category definitions

CategoryMain questionExamplesOutput
Delivery workflow/methodologyHow should humans and agents deliver software?Spec Kit, OpenSpec, AI-DLC, GSD, Superpowersspecs, plans, tasks, reviews, approvals
Coding agent harness/runtimeHow does an AI coding agent run with tools, memory, files, shell, and policies?Codex CLI, Claude Code, Hermestool calls, code changes, session memory
Agent app frameworkHow do we build an AI app or agent service?LangChain, LangGraph, OpenAI Agents SDK, AutoGen, CrewAI, Google ADKchains, graphs, agents, state, tool calls
Managed agent platformHow do we host/manage enterprise agents with provider services?Azure AI Foundry Agent Service, Amazon Bedrock Agentsmanaged agents, connectors, deployment controls
Low-code AI workflowHow do non-specialist teams compose AI workflows quickly?Dify, n8nworkflow canvas, nodes, app templates
Data/RAG frameworkHow do we ingest, index, retrieve, and ground knowledge?LlamaIndex, Haystackindexes, retrievers, document pipelines
Evals/observabilityHow do we know behavior is correct in production?LangSmith, Langfuse, Phoenix, OpenTelemetrytraces, metrics, eval scores, alerts

How adjacent tools relate to this guide

ToolBest mental modelCompetes withDoes not replace
OpenAI Agents SDKApp-level agent SDK for tool-using agentsOther app frameworks/SDKs in some use casesAI-DLC, Spec Kit, OpenSpec, release governance
AutoGenMulti-agent programming frameworkCrewAI, LangGraph in some multi-agent workloadsDelivery workflow, source-of-truth artifacts
CrewAIRole/task oriented multi-agent frameworkAutoGen, LangGraph in some workloadsGovernance, evals, tool policy
Google ADKAgent development kit for building/deploying agentsOther agent app SDKsSDD, AI-DLC, coding harness
Azure AI Foundry Agent ServiceManaged enterprise agent serviceOther managed agent platformsRepo-level delivery workflow
Amazon Bedrock AgentsManaged AWS agent capabilityOther managed agent platformsSpec discipline or AI-DLC decisions
DifyLow-code LLM app/agent workflow platformn8n, some app framework use casesDeep codebase workflow governance
n8n AI AgentWorkflow automation with AI agent nodesDify, automation platformsSpec, tests, code review discipline
LlamaIndexData/RAG-focused frameworkLangChain for data-heavy RAG flowsDelivery workflow or agent harness
Semantic KernelApp orchestration SDK with enterprise integration patternsLangChain/OpenAI SDK in some contextsAI-DLC governance

When to add a deep dive

Do not add a deep dive for every popular AI tool. Add a deep dive only when one of these is true:

TriggerAdd a page?Why
Users commonly confuse it with SDD/workflow frameworksYesIt reduces decision confusion
It changes source-of-truth artifactsYesIt affects delivery method
It is only an implementation libraryMaybeMention in ecosystem map unless heavily requested
It is provider-managed and enterprise-specificMaybeAdd a platform page if readers need deployment guidance
It is low-code and audience is non-developerMaybeAdd only if the guide targets citizen developers too

Practical selection examples

SituationRecommended combination
Build a custom RAG backend in codeOpenSpec + LangChain or LlamaIndex + evals
Build a stateful support agentAI-DLC or OpenSpec + LangGraph + tool policy
Build a multi-agent research prototypeOpenSpec + AutoGen or CrewAI + Superpowers
Build enterprise agent on AzureAI-DLC + Azure AI Foundry Agent Service + eval/audit gates
Build enterprise agent on AWSAI-DLC + Amazon Bedrock Agents + eval/audit gates
Build internal automation with low codeOpenSpec-lite + n8n or Dify + tool permission matrix
Build your own coding/research agent harnessHermes + Superpowers + tool permission matrix

Official references

Bottom line

If a tool builds the runtime behavior, it belongs in the app/platform layer. If it governs how code changes are specified, approved, implemented, and reviewed, it belongs in the workflow layer. Most strong teams need both, but they should not confuse the two.

Built as a static bilingual AI engineering stack guide.