Agents Need a Foundation

Today I still see people talking about agentic things, agentic approaches, and a lot of agent-related ideas, but sometimes I feel they do not know the context. I have been following this topic for a while, agentic development and engineering, and the work happening around LangGraph and LangChain. I have watched companies and enterprises move to LangGraph, and that is totally fine.

But people keep selling courses about harnesses and agentic things, and a lot of them do not really know why any of it should be applied. It feels like the foundation is missing. Sometimes, when we set out to build something agentic, we end up reinventing the same LangGraph and LangChain thinking anyway. What those teams have learned and actually built is genuinely important for us.

At the same time, if we are building for a local context, some of those learnings and architectures simply do not fit. We need different approaches. So I think we should be more creative with the thinking we have right now, because there are many context issues and memory issues, and we are still experimenting to find what works best for different scenarios.

We should be more open about all of this, rather than treating it like learning C++ or Java. This is very experimental. We can be more creative and think harder. We can go deeper and deeper, instead of just making slop.