The Profession

The preceding chapters developed MAGE from an engineering problem into a method and then a theory. Commodity intelligence makes implementation abundant relative to engineering judgment. Modeling makes consequential knowledge and intent explicit so that engineers and agents can reason about systems without repeatedly reconstructing them from implementation. Alignment makes selected obligations enforceable through the surrounding engineering environment. Governance conversion turns recurring failures and judgment into durable structure that subsequent work can inherit. Chapter 6 examined the resulting dynamics and economics: when these structures pay for themselves, how they change the division of engineering work, and what the theory predicts.

Together, these ideas suggest a change in where engineers exercise their expertise. As agents assume more implementation work, engineers can increasingly direct that work through representations, evidence, constraints, and controls rather than through the implementation itself. This does not mean that every engineer becomes a modeler, or that every consequential judgment can be mechanized. It means that engineers have new ways to exercise informed control over systems whose implementation they increasingly delegate. Which ways prove economical, and which judgments remain with people, are engineering questions in their own right.

This final chapter considers what follows for the profession. It begins with the reorganization of software work, places MAGE within established engineering traditions, examines what transfers beyond software, and concludes with the engineer’s expertise, authority, education, and responsibility.

Key Ideas

When implementation becomes abundant, software engineering does not disappear. Its scarce work moves. Engineers can delegate more realization while remaining responsible for the resulting system. Their work increasingly centers on retaining informed control: deciding what must be true, choosing representations that make consequential properties understandable, establishing evidence, allocating authority, coordinating autonomous work, and accepting consequential tradeoffs.

The same account suggests a broader engineering problem: how people retain informed control when agents perform consequential engineering work in domains whose representations, evidence, and means of enforcement differ from software.

© James C. Davis, 2026–present