I’ve been building with LangGraph for the past few months, and honestly? I’m starting to wonder if we even need it anymore. 🤔 Here’s what happened: I was about to refactor a multi-agent system using LangGraph when I realized something. The landscape has completely changed in the past year.
The setup used to be:
- Build agents in LangGraph
- Wire them together with custom state management
- Hope they play nice with each other 🤞
- Pray it works in production
The setup now:
- Any LLM SDK (OpenAI, Anthropic, whatever)
- Standard protocols for connecting agents to tools
- Standard protocols for agents talking to each other
- Done ✅
What’s wild is how fast this happened. Major tech companies are now backing open standards for agent communication. Meanwhile, framework-specific solutions feel increasingly… niche.
Here’s the thing though 💡 : LangGraph solved a real problem when it launched in early 2024. Multi-agent coordination was a mess. But that was before standardization kicked in. Now? If I build an agent with one SDK and you build yours with another, we can make them talk through open protocols. No framework lock-in. No proprietary structures. Just standards.
My experience building with LangGraph 🤖 : The agents felt… robotic. Too much hardcoded logic. Too many predefined nodes and edges. The LLM’s natural intelligence gets boxed into rigid graph structures, and suddenly your “intelligent agent” is just following a flowchart. When I let the LLM decide naturally (with proper tool calling), the conversations flow better. The agent feels smarter. Less “if this then that,” more actual reasoning.
The reality check ✅ : For simple tool-calling, you never needed LangGraph. The LLM SDK handles it. For complex workflows within a single agent, maybe LangGraph still makes sense. But for multi-agent systems? Open protocols are emerging as the way forward. 🚀 So where does that leave frameworks like LangGraph? Honestly, I’m not sure anymore.
Maybe I’m missing something though. 🤷♂️ For those of you building multi-agent systems right now: are you still using orchestration frameworks like LangGraph? Or have you moved to protocol-based approaches? And if you’re sticking with frameworks, what’s the reason? I’m genuinely curious because I might be about to rewrite a bunch of code, and I want to make sure I’m not throwing out something valuable. Drop your thoughts below 👇
P.S. If you’re exploring alternatives, it’s worth looking at what open standards are emerging in this space. The shift is happening faster than I expected. ⚡