How I Use AI in Engineering

AI can dramatically accelerate engineering, but speed only becomes valuable when it remains inside a disciplined loop of context, judgment, verification, and operational feedback.

AI Is Part of the Engineering Process

I use AI as an active engineering collaborator—not as a substitute for experience, accountability, or clear thinking.

It is most valuable when it helps me move faster through the parts of a problem that benefit from breadth: exploring alternatives, challenging assumptions, organizing unfamiliar information, generating test cases, and turning rough ideas into something concrete enough to evaluate.

The distinction matters. Producing an answer is not the same as producing a working system.

Software still has to operate in a real environment, with real users, failure modes, integrations, budgets, and consequences. The engineer remains responsible for deciding what should be built, whether an answer is correct, and whether the result is safe enough to put into production.

Start With the Problem, Not the Prompt

Good AI-assisted work begins before the model is involved.

I first establish the actual business or operational problem, the constraints around it, and what success needs to look like. That context determines whether AI is useful at all—and, if it is, what role it should play.

I then give the model enough relevant material to reason with:

• Existing architecture and interfaces
• Business and operational rules
• Representative examples
• Existing code and documentation
• Test results
• Known constraints and failure conditions

The quality of the collaboration is largely determined by the quality of this context.

A Controlled Loop

My working process is iterative. AI can propose, explain, compare, and implement. I review, test, redirect, and decide.

When evidence exposes a weak assumption, the work moves backward rather than forcing an attractive answer forward.

The diagram is intentionally a loop. Production is not the end of reasoning. Operating software creates new evidence, and that evidence often changes the next version of the problem.

Where AI Earns Its Place

AI is especially effective for:

• Accelerating technical reconnaissance
• Comparing architectural approaches
• Creating an initial implementation
• Finding edge cases
• Explaining unfamiliar code
• Producing migrations and test scaffolding
• Documenting decisions while they are still fresh

It also gives me a tireless second perspective. I can ask it to argue against a design, identify hidden coupling, enumerate failure paths, or explain why an apparently simple change may be dangerous.

That does not make its critique automatically correct. It makes the critique available early, when changing direction is inexpensive.

Verification Is Not Optional

AI output is a hypothesis until it has been verified.

Code must compile. Tests must exercise meaningful behavior. Security boundaries must be examined. Claims about a library, protocol, or platform must be checked against authoritative documentation. Generated solutions must fit the system that actually exists, not an idealized version inferred from a few files.

This is where experience becomes more important, not less.

Faster production increases the number of decisions that can be made in a day. It does not reduce the need to recognize which decisions carry risk.

Human Judgment Remains the Gate

I do not ask AI to assume responsibility for a production decision. It cannot understand organizational risk, customer expectations, operational history, or the consequences of failure in the same way that the people responsible for the system do.

AI can inform the decision. It can improve the implementation. It can expose considerations that might otherwise be missed.

The final judgment still belongs to the engineer.

The Result I Am After

My goal is not to maximize the amount of AI-generated code. It is to shorten the distance between a difficult problem and a dependable solution.

Used well, AI reduces mechanical effort and expands the number of approaches that can be considered. That creates more room for the work that matters most:

• Understanding the operation
• Choosing the right system boundaries
• Testing the uncomfortable cases
• Measuring the result
• Accepting responsibility for what reaches production

AI changes the pace of engineering. Judgment still determines its direction.