Skip to content

Engineering & GenAI

Premise. Commodity intelligence changes the economics of software engineering. As implementation becomes abundant, engineering does not disappear: its scarce resources become more visible.

This is the course's opening argument. It asks what changes when implementation capacity becomes abundant, and why generative AI is an engineering tool to be governed rather than a replacement for engineering judgment. The rest of the course develops the response.

Six claims about software engineering with GenAI

This unit introduces six claims about how GenAI changes software engineering:

  1. Commodity intelligence changes the economics of software engineering. Implementation capacity is becoming abundant relative to engineering judgment. As implementation gets cheaper, engineering effort shifts toward what remains scarce: deciding what to build, representing the system clearly enough to reason about it, producing evidence, and making important requirements enforceable by the engineering environment.
  2. Scale creates a reasoning problem. Large software systems already exceed the reasoning horizon of humans; agents inherit the same problem. Software engineering has always answered scale with abstraction. Commodity intelligence does not remove that need. It makes the representations that guide the work more important.
  3. Modeling makes engineering knowledge and intent explicit. Purposeful models capture the knowledge and intent needed for engineering decisions while leaving irrelevant choices open. As commodity intelligence lowers the cost of deriving, maintaining, and using such representations, more engineering knowledge can be carried forward rather than reconstructed from implementation. Models let humans and agents reason about larger properties while deliberately leaving realization choices open where engineering has imposed no obligation.
  4. Alignment makes engineering obligations enforceable. Important engineering decisions cannot live only in instructions to an agent or in a person's head. Alignment encodes selected obligations into checks and controls that can constrain work or determine what the environment will accept. Engineers can then give agents substantial freedom in how they build the system while retaining control over the properties that matter.
  5. Governance conversion turns recurring judgment into durable engineering structure. When a failure exposes missing knowledge or an unenforced obligation, encode the lesson into a model, procedure, or mechanism that future work can inherit. Durable structure becomes engineering capital when later work keeps benefiting from it.
  6. Engineering work will reorganize around what remains scarce. As implementation becomes cheaper, more engineering effort will move toward representation, evidence, governance, coordination, and judgment. Agents may perform increasing portions of that work as well. The durable boundary is responsibility for deciding what matters, what evidence is sufficient, which obligations should be enforced, and what tradeoffs remain acceptable.

Measurement for decision-making

Each unit of this course develops a model and then asks the same question of it: what could we observe that would tell us whether the model still holds, and whether the decision it produced should stand? Here the question is are we keeping informed control over consequential change? Changeability creates leverage only while engineers can still use it deliberately, so observe what change costs: how long a consequential change takes, how far a modification propagates, how often a change introduces a failure, how long it takes to obtain evidence that a change is acceptable. None of these measures engineering quality in general. A system meant to stay changeable that now demands broad, risky, or expensive modification is evidence about the structures governing change, not about the most recent change.

From the claims to MAGE

These six claims lead to MAGE: Model-Based Agentic Engineering. Its working cycle is: model consequential knowledge; enforce important obligations; do the governed work; convert recurring failures and judgment into durable structure; repeat. The goal is not maximum automation. It is to make greater autonomy possible while preserving the engineering decisions and controls that matter.

One question recurs throughout the course: How do we safely grant autonomy to commodity intelligence—and what cannot be delegated? The unit develops Modeling, Alignment, and governance conversion from this engineering problem rather than presenting MAGE as a collection of prescribed practices.

Read the expanded treatment: The Software Engineering Handbook, "Software Engineering" →

Materials

  • Lecture slides — GenAI as an engineering tool — source (PPTX)

Readings

The new engineering problem

  • MAGE Part I, "The New Engineering Problem." Develops the premise that commodity intelligence changes the economics of software engineering by making implementation capacity abundant relative to engineering judgment. Introduces the resulting imbalance and asks where engineering effort moves when producing implementation is no longer the dominant constraint. Full citation: James C. Davis, Model-Based Agentic Engineering, 1st ed. (2026), https://davisjam.github.io/model-based-agentic-software-engineering/book/mage-book/index.html, Part I, "The New Engineering Problem."
  • Brooks, "No Silver Bullet." The classic statement that software's essential difficulty is conceptual — deciding what to build and keeping it coherent — while tools address only accidental complexity. It frames why making implementation cheaper moves the engineering bottleneck rather than removing it, the same claim GenAI now tests at scale. Full citation: Frederick P. Brooks, “No Silver Bullet: Essence and Accidents of Software Engineering,” Computer 20, no. 4 (1987): 10–19.

The MAGE argument