§1.1 The Printer

Coding agents make software implementation cheap enough to change the engineering problem. Describe a change and an agent can often implement it quickly. The opportunity is substantial, but what should we do with this new capacity?

For an analogy, let us begin with office supplies. A stapler joins sheets of paper and a hole punch makes holes in them. Their purposes and limitations are in their names. A printer leaves more for us to decide. The same 2D printer can produce a newspaper, a poster, or a map, each with its own conventions for layout and legibility. A 3D printer can turn a description of a shape into a mounting bracket, a protective enclosure, or an adapter, each with different requirements for strength, clearance, and fit. As people discover what the machinery can usefully make, they develop templates, reusable designs, and production practices that give the next person a place to start.

Software agents are printers, not staplers. They likewise supply broad productive capacity. Like printers, agents can follow instructions. But agents can infer much more from an incomplete description than a physical printer can: an algorithm, a data structure, or a decomposition may never have been specified, and yet an agent can choose them and produce an implementation. That inferential capacity is precisely what makes the machinery powerful.

When we work with agents, we delegate some judgment as well as production. We could specify the entire implementation, hand it to an agent, and ask it to reproduce it. That would be an unusually expensive way to copy a file. Much of the opportunity lies in letting agents work out things we have not already worked out ourselves. The engineering question is which decisions we should make explicit and which we can responsibly leave to the agent.

© James C. Davis, 2026–present