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
| Category | Main question | Examples | Output |
|---|---|---|---|
| Delivery workflow/methodology | How should humans and agents deliver software? | Spec Kit, OpenSpec, AI-DLC, GSD, Superpowers | specs, plans, tasks, reviews, approvals |
| Coding agent harness/runtime | How does an AI coding agent run with tools, memory, files, shell, and policies? | Codex CLI, Claude Code, Hermes | tool calls, code changes, session memory |
| Agent app framework | How do we build an AI app or agent service? | LangChain, LangGraph, OpenAI Agents SDK, AutoGen, CrewAI, Google ADK | chains, graphs, agents, state, tool calls |
| Managed agent platform | How do we host/manage enterprise agents with provider services? | Azure AI Foundry Agent Service, Amazon Bedrock Agents | managed agents, connectors, deployment controls |
| Low-code AI workflow | How do non-specialist teams compose AI workflows quickly? | Dify, n8n | workflow canvas, nodes, app templates |
| Data/RAG framework | How do we ingest, index, retrieve, and ground knowledge? | LlamaIndex, Haystack | indexes, retrievers, document pipelines |
| Evals/observability | How do we know behavior is correct in production? | LangSmith, Langfuse, Phoenix, OpenTelemetry | traces, metrics, eval scores, alerts |
How adjacent tools relate to this guide
| Tool | Best mental model | Competes with | Does not replace |
|---|---|---|---|
| OpenAI Agents SDK | App-level agent SDK for tool-using agents | Other app frameworks/SDKs in some use cases | AI-DLC, Spec Kit, OpenSpec, release governance |
| AutoGen | Multi-agent programming framework | CrewAI, LangGraph in some multi-agent workloads | Delivery workflow, source-of-truth artifacts |
| CrewAI | Role/task oriented multi-agent framework | AutoGen, LangGraph in some workloads | Governance, evals, tool policy |
| Google ADK | Agent development kit for building/deploying agents | Other agent app SDKs | SDD, AI-DLC, coding harness |
| Azure AI Foundry Agent Service | Managed enterprise agent service | Other managed agent platforms | Repo-level delivery workflow |
| Amazon Bedrock Agents | Managed AWS agent capability | Other managed agent platforms | Spec discipline or AI-DLC decisions |
| Dify | Low-code LLM app/agent workflow platform | n8n, some app framework use cases | Deep codebase workflow governance |
| n8n AI Agent | Workflow automation with AI agent nodes | Dify, automation platforms | Spec, tests, code review discipline |
| LlamaIndex | Data/RAG-focused framework | LangChain for data-heavy RAG flows | Delivery workflow or agent harness |
| Semantic Kernel | App orchestration SDK with enterprise integration patterns | LangChain/OpenAI SDK in some contexts | AI-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:
| Trigger | Add a page? | Why |
|---|---|---|
| Users commonly confuse it with SDD/workflow frameworks | Yes | It reduces decision confusion |
| It changes source-of-truth artifacts | Yes | It affects delivery method |
| It is only an implementation library | Maybe | Mention in ecosystem map unless heavily requested |
| It is provider-managed and enterprise-specific | Maybe | Add a platform page if readers need deployment guidance |
| It is low-code and audience is non-developer | Maybe | Add only if the guide targets citizen developers too |
Practical selection examples
| Situation | Recommended combination |
|---|---|
| Build a custom RAG backend in code | OpenSpec + LangChain or LlamaIndex + evals |
| Build a stateful support agent | AI-DLC or OpenSpec + LangGraph + tool policy |
| Build a multi-agent research prototype | OpenSpec + AutoGen or CrewAI + Superpowers |
| Build enterprise agent on Azure | AI-DLC + Azure AI Foundry Agent Service + eval/audit gates |
| Build enterprise agent on AWS | AI-DLC + Amazon Bedrock Agents + eval/audit gates |
| Build internal automation with low code | OpenSpec-lite + n8n or Dify + tool permission matrix |
| Build your own coding/research agent harness | Hermes + Superpowers + tool permission matrix |
Official references
- OpenAI Agents SDK: https://platform.openai.com/docs/guides/agents-sdk/
- Microsoft AutoGen: https://microsoft.github.io/autogen/
- CrewAI: https://docs.crewai.com/
- Google Agent Development Kit: https://google.github.io/adk-docs/
- Azure AI Foundry Agent Service: https://learn.microsoft.com/azure/ai-foundry/agents/overview
- Amazon Bedrock Agents: https://docs.aws.amazon.com/bedrock/latest/userguide/agents.html
- Dify: https://docs.dify.ai/
- n8n AI Agent node: https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.agent/
- LlamaIndex: https://docs.llamaindex.ai/
- Haystack: https://docs.haystack.deepset.ai/
- Semantic Kernel: https://learn.microsoft.com/semantic-kernel/
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.