Introducing Space Engineering: A New Definition of AI Engineering

Over the past few years, anyone building AI applications has seen a succession of newly coined concepts:

Prompt Engineering, Context Engineering, Harness Engineering, Loop Engineering, Graph Engineering.

Now I want to propose another one: Space Engineering.

I increasingly believe that the essence of human–AI interaction is not simply writing prompts, supplying context, calling tools, or designing workflows.

At a deeper level, we are defining a search space for the model.

The prompt influences where it looks. Context determines what it can see. Tools determine what it can do. Workflows determine which paths it can take. Evaluation determines how it knows whether it is moving in the right direction.

Together, these elements form the space in which the model solves the problem.

If the space is defined incorrectly, even the strongest model will only search within the wrong boundaries.

If the space is too large, the model drifts and struggles to find the best path.

If the space is too narrow, it loses the freedom required to discover a better solution.

Good AI engineering is therefore not about telling the model every step to take. It is about designing the objective, boundaries, constraints, tools, and feedback so clearly that the right answer becomes the easiest one to reach.

From this perspective, Prompt Engineering, Context Engineering, Harness Engineering, and the rest are all local forms of Space Engineering.

The core capability in AI application development is not learning how to “talk to the model” more effectively.

It is learning how to design the space in which the model searches for a solution.