Context Engineering Is the New Prompt Engineering

"Prompt engineering" had a good run as the headline skill of the AI era. It's already being replaced by something more durable and more technical: context engineering — the practice of deciding what information a model gets to work with before it generates a single token.

Prompts were never the bottleneck

A well-phrased request helps at the margins. But the difference between a useless answer and a brilliant one is almost never the wording — it's whether the model could see the relevant code, the existing patterns, the constraints, and the definition of done. Give a mediocre prompt rich context and you get gold. Give a beautiful prompt no context and you get confident fiction.

What good context looks like

For coding work, good context is the surrounding module, the test that specifies the behavior, the types, and one or two examples of how your team does this kind of thing already. For agents, it extends to the right tools and the right files — and, crucially, *not* the wrong ones. Relevance beats volume every time.

Curate, don't dump

The instinct to paste everything is the enemy. Models have finite attention, and noise actively degrades output — a giant blob of unrelated code makes the model *less* likely to follow the pattern that matters. The work is editorial: select the few things that matter, omit the rest, and structure it so the signal is obvious. That's a real engineering skill, and it's the one that's compounding in value.