Decision Guide

The wrong question is "Which framework is best?" The useful question is "Which layer of AI engineering is failing in my context?"

First question	Go to
Are you building an AI app?	LangChain, LangGraph, Data/RAG, Tools/MCP
Are you improving software delivery with agents?	Spec Kit, OpenSpec, AI-DLC, GSD, Superpowers
Are you running/customizing an agent harness?	Hermes, Codex vs Claude vs Hermes
Are you preparing for production?	Evals & Observability, Security & Governance
Are you choosing models or serving strategy?	Model & Serving Layer

flowchart TB
    Q["What are you building or choosing?"] --> A0{"AI app / agent backend?"}
    A0 -->|Yes| A1{"Simple chains, RAG, tools?"}
    A1 -->|Yes| LC["LangChain"]
    A1 -->|No, stateful graph| LG["LangGraph"]
    A0 -->|No| R{"Runtime or workflow?"}
    R -->|Runtime/harness| H{"Need open-source custom runtime?"}
    H -->|Yes| HA["Hermes Agent"]
    H -->|No| HC["Codex CLI or Claude Code"]
    R -->|Workflow/methodology| A{"Governance, audit, many stakeholders?"}
    A -->|Yes| AIDLC["AWS AI-DLC Workflows"]
    A -->|No| B{"Are requirements/specs vague?"}
    B -->|Yes| SK["GitHub Spec Kit"]
    B -->|No| O{"Want lightweight iterative change specs?"}
    O -->|Yes| OS["OpenSpec"]
    O -->|No| C{"Long project, many phases, context breaks?"}
    C -->|Yes| GSD["GSD"]
    C -->|No| D{"Need TDD, review, agent discipline?"}
    D -->|Yes| SP["Superpowers"]
    D -->|No| MIN["Minimal workflow: clarify -> implement -> test -> review"]

Choose by need

I need...	Choose
Full AI engineering stack context	AI Engineering Stack Map
Model routing, local LLMs, or serving strategy	Model & Serving Layer
RAG data pipeline and retrieval quality	Data, RAG & Retrieval
Safe tool use, MCP, or tool gateways	Tools, MCP & Gateways
Evals, tracing, and production feedback	Evals & Observability
Security, governance, and risk tiers	Security & Governance
Build AI app, RAG, or tool-calling agent	LangChain
Build stateful long-running agent backend	LangGraph
A polished coding agent CLI	Codex CLI or Claude Code
An open-source/customizable agent runtime	Hermes Agent
A way to describe features so AI builds the right thing	Spec Kit
A lightweight spec layer for iterative brownfield changes	OpenSpec
A lifecycle with approval and audit	AWS AI-DLC Workflows
A system for many phases across many sessions	GSD
A skill layer that stops the agent from coding recklessly	Superpowers
MVP speed with some structure	GSD + lightweight Spec Kit
Important product feature with acceptance criteria	Spec Kit + Superpowers
Enterprise modernization	AWS AI-DLC primary
Safer refactoring	Superpowers + tests
Compliance or security-sensitive delivery	AWS AI-DLC + explicit security gates

Choose by work size

Work size	Recommended workflow
5-30 minutes	Superpowers light or manual prompt with tests
Half day to 2 days	Spec Kit or Superpowers
1-5 day iterative brownfield change	OpenSpec
1-3 weeks	Spec Kit + GSD, or AWS AI-DLC if risk is high
1-3 months	AWS AI-DLC or GSD with added governance
Enterprise program	AWS AI-DLC primary; others as supporting layers

Choose by codebase

Codebase	Best fit
Greenfield product app	Spec Kit for clarity; GSD for speed
Brownfield monolith	AWS AI-DLC for modernization; Superpowers for targeted refactors
Brownfield feature change with low-medium risk	OpenSpec
API/library	Spec Kit because contract clarity matters
Internal tool	GSD or Spec Kit
Regulated system	AWS AI-DLC
Open source project	Spec Kit or Superpowers for PR reviewability

Red flags

Red flag	What to avoid
You want to install all frameworks at once	You will create multiple sources of truth
You cannot review generated docs	Do not use a heavy governance flow
Your CI is weak	Do not allow broad automated execution
The feature is security-sensitive	Do not use a purely speed-optimized workflow
The task is tiny	Do not create full lifecycle artifacts
You need formal audit	Do not rely on OpenSpec alone
You only need a workflow process	Do not add Hermes unless runtime customization matters