Skip to content
TO
All news

· 2 min read

Frameworks make you productive. Fundamentals make you dangerous.

I spent two months building an AI agent's memory, retries and orchestration by hand before discovering LangGraph. Was it wasted time?

  • ai
  • agents
  • engineering

The software engineering landscape has changed dramatically. Someone can now go from zero to an impressive application without deeply understanding what's happening under the hood, and the conversation has shifted from "How did you build it?" to "Does it work?"

I still believe there's tremendous value in building things from scratch, but it depends on the stage you're in. If you're learning, build from scratch. Break things. Reinvent the wheel. That's where real understanding comes from. Once you understand the fundamentals, speed becomes a competitive advantage, and leveraging existing tools isn't laziness. It's engineering.

Two months with nothing but Python

When I built my AI Sales & Support Agent, I started with nothing but Python and determination. For nearly two months I built everything myself:

  • Retry logic and exponential backoff
  • Conversation memory and long-term memory
  • RAG retrieval logic and memory loading strategies
  • Conversation summarisation to reduce token usage
  • Tool orchestration and state management

It was frustrating, but incredibly educational.

Later, in an interview, I explained how the system worked. The interviewer asked, "Why didn't you use LangGraph or LangChain?" At the time I had never used either. Looking back, that was probably one of the reasons I wasn't selected.

What I found when I tried them

When I eventually discovered LangGraph and LangChain, I had one overwhelming realisation: these frameworks had already solved nearly every problem I'd spent weeks solving.

So did I waste two months? I don't think so. Because I built everything from first principles, I now understand why LangGraph works, not just how to use it. That gives me the confidence to debug it, customise it, and replace parts of it when I need to.

The takeaway

In a world where AI makes software development more accessible, the gap between developers isn't only about who ships fastest. It's about who understands the systems they're building, can adapt them when the need arises, and knows why they're building them.

Frameworks make you productive. Fundamentals make you dangerous.

Would I build everything from scratch again? Probably not. Am I glad I did it once? Absolutely.

Tabot Oben

Full-Stack AI Engineer in Douala, Cameroon. Get in touch.