Open-source agent quad: LangGraph + LlamaIndex + CrewAI + AutoGen — four 1.0-class milestones in one week, four different agent models

5 sources

LangGraph 0.5 (graph-based orchestration with persistent state, human-in-the-loop interrupt/resume, time-travel debugging), LlamaIndex Workflows GA (event-driven orchestration with step-level async / sync / streaming handlers), CrewAI Flows GA (event-driven orchestration with persistent state across crews + multi-agent crews with role / goal / backstory primitives), and AutoGen 0.4 (actor-model architecture replacing the group-chat manager abstraction) form the open-source agent framework quad. Each framework takes a different bet on what an agent is: LangGraph on graph state, LlamaIndex on event-driven steps with RAG primitives, CrewAI on role-based multi-agent collaboration, AutoGen on actor-message passing. Together they show that open-source agent runtimes have stabilized around four distinct models.

Why now

All four shipped 1.0-class milestones in the same week — the open-source agent runtime quad is now stable, and creators can pick the framework that matches their mental model.

Why it is worth publishing

Demo potential: side-by-side of the same agent workflow built on each of the four frameworks, measuring developer ergonomics and observability surface.

Evidence basis

Four independent framework primary sources + The Decoder weekly roundup

“Four open-source agent frameworks shipped 1.0-class milestones in one week — LangGraph, LlamaIndex, CrewAI, AutoGen — and each one takes a different bet on what an agent is: graph state, event-driven steps, role-based collaboration, actor-message passing.”

Angle

Use the quad update to introduce the 'four different agent models' framing — each open-source framework takes a different bet on what an agent is — and use the bet comparison as the lens for picking a framework.

Format

Long-form explainer

Demo idea

Record a 12-minute explainer: 3 min intro on 'four agent models' framing, 2 min per framework (LangGraph / LlamaIndex / CrewAI / AutoGen), 2 min on a live side-by-side of the same agent workflow on all four.

Platform notes

Specific per-node execution semantics, per-step instrumentation hooks, per-flow state primitives, and actor-message-passing semantics beyond the captured summary were not extracted; confirm specific primitives against the underlying framework docs before stating them on the record. AutoGen cross-language support beyond Python + .NET is not extracted.

Usable claims

LangChain shipped LangGraph 0.5 framework (graph-based agent orchestration with persistent state, human-in-the-loop interrupt/resume, time-travel debugging) + LangGraph Studio GA.
LlamaIndex shipped Workflows GA (event-driven agent orchestration with step-level async / sync / streaming handlers, automatic instrumentation hooks) + LlamaDeploy 1.0.
CrewAI shipped Flows GA (event-driven orchestration with persistent state across crews) + Crews 1.0 (multi-agent crews with role / goal / backstory primitives).
Microsoft shipped AutoGen 0.4 — actor-model architecture refactor (replacing the group-chat manager abstraction), async-first design, cross-language support (Python + .NET).

Evidence pipeline

From the news

Breakdown

LangGraph, LlamaIndex, CrewAI, and AutoGen all shipped 1.0-class milestones, but each takes a different bet on what an agent is. This explainer keeps the four bets distinct (graph state, event-driven steps, role-based collaboration, actor-message passing) rather than treating 'open-source agent framework' as a single category.

Sources

Risks

Docs confirm the framework features exist but specific primitive signatures and protocol details beyond the captured summary are not in this pass. Verify specific capability claim against the underlying vendor docs and the actual license / pricing matrix before stating it on the record; do not paraphrase per-platform pricing or license terms into specific dollar figures or commercial-use clauses.
AutoGen docs confirm Python + .NET but additional language support beyond the captured summary is not in this pass. Verify specific capability claim against the underlying vendor docs and the actual license / pricing matrix before stating it on the record; do not paraphrase per-platform pricing or license terms into specific dollar figures or commercial-use clauses.
Use The Decoder and IT之家 as media-type corroboration, but read the underlying framework docs for any specific primitive or API surface before stating it on the record. Verify specific capability claim against the underlying vendor docs and the actual license / pricing matrix before stating it on the record; do not paraphrase per-platform pricing or license terms into specific dollar figures or commercial-use clauses.

Demo ideas

Side-by-side agent workflow: same recurring task on each of the four frameworks, measure developer ergonomics and observability surface.
Bet matrix: plot each framework on 'what an agent is' dimensions (state model, execution model, multi-agent primitive, observability surface).
Migration story: 'migrate an agent from one framework to another', measure porting work and surface delta.