Conclusion
Though much is taken, much abides.
— Alfred, Lord Tennyson, Ulysses
The question I asked
In the Preface, I asked:
How do we safely grant autonomy to commodity intelligence?
Software engineering provided a concrete setting for the question: if an agent can perform consequential work at extraordinary speed, what must the surrounding environment provide before we can safely grant that capability useful autonomy?
The seven chapters developed an answer at different levels. Modeling established how to represent consequential engineering knowledge; Alignment showed how to enforce selected obligations independently of the producing agent. MAGE in Motion developed the method through which experience becomes durable engineering structure. DocAble Among Other Software Factories examined that method in production, while The Theory explained its mechanisms and economic consequences. Finally, The Profession considered what changes for engineers when implementation can increasingly be delegated.
I developed this answer through software engineering because that is where my engineering expertise lies. The underlying question reaches beyond software and perhaps beyond the conventional engineering disciplines. Its answer in another domain will depend on the representations, evidence, and controls available there.
The core of MAGE
MAGE makes three engineering moves.
- Work at the right representation. Choose the reduction that makes the engineering question answerable without carrying unnecessary detail. A purposeful model can expose properties that humans and agents would otherwise have to reconstruct from implementation, making them available for analysis, review, and evidence. Build the model when those benefits repay its cost, and maintain the correspondences on which its use depends.
- Put enforcement where the obligation can be represented and decided adequately. Guidance influences behavior; enforcement determines which actions or results the environment will accept. Some obligations can be enforced directly through types or permissions that restrict actions or artifacts. Where tests or validators check an obligation, a gate can reject work when the check fails. Others become decidable only after a suitable representation exposes the relevant semantics. Leave consequential decisions to human judgment when the environment lacks the meaning or evidence needed to decide them adequately.
- Make useful judgment reusable. When a recurring or consequential engineering decision is worth carrying forward, convert it into a model, mechanism, architecture, procedure, or body of evidence that later work can inherit. Such structures become engineering capital when they improve later engineering work. Ratcheteering preserves the consequential concern while allowing the particular structure that carries it to be corrected, replaced, or retired.
Together, these moves produce the governed engineering environment. Models represent consequential knowledge, controls constrain actions or evaluate results, and evidence supports engineering judgments. People remain responsible for the resulting system. In software, the representations include requirements, architectures, behavioral models, dependencies, provenance, and measurements; the governance mechanisms are constraints, sensors, and gates. Other domains will supply different representations and different means of evidence and control.
That is Model-Based Agentic Engineering.
What changes next
When commodity intelligence makes realization cheaper, the cost of the surrounding engineering work becomes more important. If a software factory can produce changes faster than engineers can specify consequential intent, evaluate evidence, or maintain control over the resulting system, then improving realization further does not remove those constraints. Engineering attention moves to them.
The probability analysis in Chapter 6 distinguishes adequate encoding from reliable realization. An agent can faithfully fulfill encoded obligations even when they omit or misstate what the system needs. Better agents can help engineers improve the encoding, but reliable fulfillment alone does not establish that it is right. This leaves two engineering problems: representing consequential intent adequately without spending more on representation than it is worth, and establishing that the resulting work is acceptable through independent evidence and evaluation. Modeling and Alignment address these problems from different directions.
Neither problem has a fixed solution. As agents improve, they can help engineers discover representations, maintain correspondences, construct controls, and evaluate evidence. The engineering environment can become more capable alongside the agents it governs. The relevant question is how to allocate work among the reasoner, the environment, and human judgment as their respective capabilities and costs change.
How much specification, and of what kind?
This leaves two related questions open: how much consequential intent must engineers make explicit, and in what form? There may be different answers for realization and assurance. As agents become better at recovering familiar intent, engineers may need to say less to obtain a useful result. A prototype may be judged by the useful behavior it demonstrates; a production system must also provide evidence that its consequential obligations are satisfied. The latter requires suitable representations of consequential properties. The specification frontier therefore depends not only on what an agent can infer, but on what the engineer must be able to establish—and on which representations make that establishment economical.
Agents can also help construct and maintain those representations, analyze them, gather evidence, and build the controls that enforce selected obligations. This reduces the cost of the engineering machinery itself, potentially making richer representations and stronger assurance economical. The opportunity is not merely to realize systems more cheaply, but to improve the machinery through which engineers govern their realization.
The Authority and Responsibility of the Engineer
I would not define the engineer by a permanent collection of tasks that only humans can perform. The boundary will move as capabilities improve. Engineering competence will change with it, as it has changed with earlier technologies. What persists is the need for informed control over consequential work: understanding what the engineering arrangement can establish, recognizing where its evidence is insufficient, and accepting responsibility for the decisions made under its authority.
That responsibility need not mean personally making every decision. Indeed, MAGE is an argument for moving suitable decisions and decision-making processes into engineered structures so that they can operate more reliably, consistently, and economically. The engineer's responsibility extends to the design and governance of those structures, including the judgment to revise them when their assumptions no longer hold.
The question from the Preface was how to grant autonomy safely. The answer is to engineer the conditions under which autonomy can be exercised, evaluated, and governed. As commodity intelligence grows more capable, those conditions may become richer and more automated. The responsibility for establishing them remains an engineering responsibility.
Commodity intelligence has not come to abolish engineering, but to demand it.