Three terms are now fighting for the same line on AI engineering job descriptions, and using them interchangeably is a mistake. They aren’t competing buzzwords. They’re a stack. A prompt controls one model response. A loop controls one agent’s behavior cycle. A graph controls the organization of many agents. Each layer preserves the one beneath it—a prompt doesn’t vanish inside a loop, it just stops being the thing you type by hand.
Prompt engineering is the established layer, but its defining assumption—a human judging every output—breaks under high volume or multi-step tasks. The fix isn’t a better prompt. It’s the next floor up. Loop engineering exploded in developer discourse in June 2026, framed as the shift from writing prompts to designing the cycle that prompts the agent. A widely shared post and a Claude Code team talk that same week cemented the idea. The hard part isn’t the repetition, it’s the stop condition. A loop that can’t mechanically distinguish “done” from “stuck” doesn’t fail loudly—it just keeps spending tokens.
By July 2026, attention moved to graph engineering, the newest and least settled label. It collides with an older knowledge-graph usage, making its provenance unresolved. But the underlying practice, graph-based orchestration, has a documented lineage in multi-agent systems research. Inside Anthropic, researchers found that even as they scaled to 50 sub-agents, prompt engineering remained the primary lever for fixing coordination failures—not topology changes.
So when you see these terms on a job spec, the real question isn’t which one is right. It’s what layer of abstraction the role actually operates at. The stack is real. The confusion is expensive.
