Comparative research on the AI agent framework landscape in 2026.
This repository collects technical analyses, market-oriented comparisons, and real-world use cases for teams evaluating which agentic stack to adopt for production systems.
Primary goals:
- help builders compare frameworks beyond hype
- highlight tradeoffs that matter in production
- document concrete enterprise and community use cases
- create a reusable decision asset for architects, founders, and technical leaders
Current scope:
- 1 comparative market report in Italian and English
- 3 framework deep-dive tracks with bilingual analysis and use cases
- a documentation-first structure designed to scale with additional frameworks
Language coverage:
- Italian: original long-form research
- English: curated companion editions for broader reach
docs/
comparative/
agentic-ai-frameworks-analysis-it.md
agentic-ai-frameworks-analysis-en.md
frameworks/
crewai/
langgraph/
openai-agents-sdk/
AI Agent Framework Landscape
(2026)
+----------------------+------------------------+---------------------------+
| Low-level runtimes | Agent frameworks | Managed agent platforms |
+----------------------+------------------------+---------------------------+
| LangGraph | CrewAI | Azure AI Foundry |
| Microsoft MAF | OpenAI Agents SDK | Bedrock AgentCore |
| Google ADK | Agno | Bedrock Agents |
| PydanticAI | Smolagents | Gemini Enterprise |
| LlamaIndex Workflows | Haystack | LangGraph Platform |
+----------------------+------------------------+---------------------------+
Interoperability layer: MCP for tools | A2A / agent protocols for agents
Core production concerns: memory | durability | observability | HITL
- It compares frameworks across architecture, memory, observability, HITL, ecosystem, and production maturity.
- It separates open-source runtimes from cloud-managed platforms, which is often blurred in generic market maps.
- It focuses on decision quality: when to choose a framework, when to avoid it, and what risks to expect.
- It includes both technical positioning and practical use-case patterns.
| Framework | Comparative coverage | Deep dive | Use cases | Languages |
|---|---|---|---|---|
| CrewAI | Yes | Yes | Yes | IT, EN |
| LangGraph | Yes | Yes | Yes | IT, EN |
| OpenAI Agents SDK | Yes | Yes | Yes | IT, EN |
| Microsoft Agent Framework | Yes | Not yet | Not yet | IT, EN summary in main report |
| Google ADK | Yes | Not yet | Not yet | IT, EN summary in main report |
| Haystack | Yes | Not yet | Not yet | IT, EN summary in main report |
If you are short on time:
- Start from the English comparative report.
- Pick the framework deep dive most relevant to your team.
- Use the use-case documents to evaluate fit by industry and workflow pattern.
If you are making a platform decision:
- Read the comparative report.
- Compare LangGraph and CrewAI deep dives.
- Validate assumptions against your cloud, language, and governance constraints.
The analyses prioritize:
- execution model and orchestration style
- state management and persistence
- multi-agent support
- human-in-the-loop support
- observability and debugging
- language ecosystem
- enterprise maturity
- lock-in risk
This repository is best suited for:
- AI engineers
- solution architects
- CTOs and technical founders
- innovation teams evaluating agent platforms
- consultants preparing decision memos or client recommendations
Suggested next deep dives:
- Microsoft Agent Framework
- Google ADK
- PydanticAI
- LlamaIndex Workflows
- Most documents are dated and versioned because the agent tooling landscape changes quickly.
- Quantitative claims such as GitHub stars, downloads, release status, and enterprise adoption should be periodically revalidated.
- English companion documents are adapted from the Italian research and optimized for international readability.
Current repository license: MIT.
If this repository evolves primarily as a research and documentation asset, a future move to a content-oriented license such as CC BY 4.0 may be worth considering.