Each New AI Problem Means the Last One Was Solved

A new term showed up in AI circles this year: Graph Engineering. Some steps in an Agent workflow have become routine enough to code directly, with no need to ask the model each time. It's cheaper and faster. Prompt Engineering wasn't even fully figured out when Context Engineering arrived. Context was just starting to make sense when Harness came along, then Loop, then Graph. Someone looked at this list and sighed: when does the AI skills list end? But looking back, the path is clearer. When Alexandr Wang entered Y Combinator in 2016, he wanted to build AI agents for healthcare. After two months, his co-founder said it was too early. Let's switch. Wang thought back to training models at MIT: you need three things: compute, code, and training data. The first two were findable online. The third had no good solution. He started Scale AI to have people label images and text, giving models something to learn from. Once Scale AI started addressing the data problem, models got stronger, and the next problem surfaced: how do you tell the model clearly what you need? Books and tutorials on Prompt Engineering followed. Prompting matured. More tools and memory got packed in, and the problem shifted to managing what was being fed to the model. Context Engineering appeared. Agents started actually doing things, then came managing permissions, verification, and budgets. Harness appeared. Wanting Agents to keep running on their own brought Loop. Coding the steps that didn't need the model's judgment brought Graph. Every new term signals one thing: someone solved the previous problem. Graph Engineering became a problem because Agent reasoning had reached a point where people had breathing room to ask: which steps actually don't need the model deciding? That question didn't exist five years ago, because back then Agents couldn't do much of anything. Wang said last year that even if models stopped improving today, current capabilities are enough to change the world for decades. Over those decades, the list of AI skills will keep growing. What it eventually looks like, nobody knows yet.