Prompting Will Be Dead in Six Months

Recent chatter on Twitter suggests that prompt engineering will be obsolete within six months. However, a new discipline known as graph engineering is rapidly gaining traction. Graph engineering essentially breaks down AI agents into smaller pieces called nodes, with each node being a specific job such as a researcher, writer, or fact checker.

These nodes are connected through edges, which determine the next action based on the previous node's output. For instance, if a fact checker discovers an error, the edge flags it and sends the task back to the writer node, enabling precise corrections. The completion of the fact-checker's approval allows the work to proceed.

I, like many others, have been utilizing a similar orchestration approach with tools like LangGraph for over a year, albeit with a minor enhancement. The term "graph engineering" has only recently become viral, but it is essentially an evolution of previous concepts such as loop engineering, harness engineering, and context engineering. The core idea remains consistent, with slight improvements and variations over time.

With that said, prompting is not dying in six months. Rather, the notion that one can merely attach a model to a clever prompt to create a fully functional agentic system is fading. It is crucial to recognize that the true demise lies in the belief that a simple prompt can suffice for complex tasks. Keep an eye out for updates as I continue to provide insights on this evolving field.