From prompts to graphs, five shifts form one continuous line

The core skill of AI engineering has moved from prompt (2022) → context (2024) → harness (late 2025) → loop → graph engineering, and each stage was born from the limits of the one before it. As models got smarter, prompt craft mattered less; putting models into production raised hard problems — context window limits, statelessness and context rot — that gave rise to context engineering; and once file systems, sandboxes, observability and subagents became necessary, the idea of the harness, meaning everything except the model, took hold. A loop has five elements — trigger, goal, check, memory and stop condition — and an example that uses Claude Code's /loop to check a price every 2 hours, then send an alert and stop if it is under $200, captures this exactly. A graph is a structure that coordinates multiple agents and loops so they don't get in each other's way; a typical example is a workflow with three parallel researchers and a blind critic who hasn't seen the draft. When a new term comes along, two questions are enough to size it up: what problem did it come from, and what does it control that the previous stage could not?

This is a note: my own write-up after reading several sources. It has no source link; events checked against a source are kept as records.