Not-yet-AI engineers hold the advantage, writes 20-year SE

The author's argument draws on a direct analogy to software development: handing a programmer only a rough instruction like 'build this list screen nicely' produces something that does not match your intent — for example, sort order reversed or display behavior for large record counts unconsidered — and that is not the programmer's fault. The same dynamic, the author says, applies to AI. A request to 'make a script to aggregate logs' yields plausible code that does not match the actual log format, lacks the items you wanted aggregated, and ignores library constraints; adding the purpose, environment, and a few sample lines of the log sharply raises the chance the output works as-is. The difference is the amount of information given and the clarity of purpose, not the AI's capability.

The author also places this in a longer engineering tradition. The core of the book is the sequence engineers have long known as waterfall — decide first, then build — and the author considers what is currently called specification-driven development (SDD) to be the same thing. This is not a book about learning new technology; it is about aiming already-held skills at a new counterpart.

The stated reason not to chase MCP, custom skills, agent loops, or harnesses is threefold: those are means, not ends; the AI itself can guide you through the setup when you actually need it; and elaborate setups tend to have short lifespans, whereas basic interactions — conveying purpose, handing over materials, eliciting options, summarizing into documents, having a human verify — carry over across whichever AI or service you use. The stakes, then, hinge on whether the reader treats instruction quality and purpose definition as the bottleneck rather than tool familiarity. The author's own experience running a dozen-plus projects in half a year is the evidence offered; readers whose work involves sensitive data will note the author defers the line between shareable and non-shareable business information to a later chapter.