Early in my career, the redlines kept coming. At an industrial firm right out of school, I worked on drawing sets that could run to roughly 300 sheets. My project architect would give me a sketch laying out the building and highlight details from the firm’s detail binder. From those starting points, I assembled sets that became 2- or 3-inch thick stacks of paper. 

Then the redlining—the corrections—began. I would receive about a third of the set to rework, which could take two or three days. Then came the next third, and then the last. By the time I finished the final third, more redlines were arriving on the sheets I had revised first.

My project architect made curiosity part of the job. “If you have questions, ask,” he told me. “If you’re not annoying me with questions, you are not doing your job. The whole point of this is you learn to anticipate the redlines and correct them before they occur.”

The marks themselves didn’t bestow judgment. They simply generated questions about what I had misunderstood and whether a correction applied elsewhere. Correcting the drawing completed an immediate task: Understanding the answers changed how I approached the next round. Over the years, those questions, explanations, and opportunities to apply what I learned helped develop judgment.

Later, the expectation changed; being a good drafter meant checking my own work. Moving toward becoming an architect meant growing my capability of checking someone else’s work. Examining my own work required understanding how and why something was wrong, not merely recognizing that it differed from something I had drawn previously.

Some of this progression established technical competence: understanding a requirement and applying it correctly. Judgment built on that foundation but extended beyond it. When information and competing priorities did not produce an obvious answer, I needed to understand why one defensible choice should prevail over another, not simply whether a drawing was technically correct.

This distinction matters in the AI office. My most recent essay on Common Edge, “How Does Knowledge Get Pass On in the AI Office,” examined how automation might remove both entry-level employment and the learning embedded in project assignments. But what should replace that experience? One possible solution is to have software produce the initial material and assign beginners to review and revise it. What prepares someone to recognize a faulty assumption or an unresolved relationship if they have never learned to investigate the problem? Checking is itself a capability that must be learned.

In a recent Randy Deitsch essay, “If Everybody’s Using AI, What’s An AI Design Studio For?,” the author asks whether AI-assisted student work demonstrates judgment or merely tool fluency. Firms inherit that assessment problem. A convincing drawing does not tell a supervisor what the beginner understands. But answering that problem does not require preserving every repetitive task. It requires distinguishing disposable production from foundational learning—and that distinction depends on the learner. 

Producing every line myself wasn’t the essential lesson. What mattered was working through relationships, encountering gaps in my understanding, and having someone help me investigate them.

There was nothing inherently educational about the size of my drawing sets. Nor was producing every line myself the essential lesson. What mattered was working through relationships, encountering gaps in my understanding, and having someone help me investigate them. AI can support that process. The essential activity is examining the reasoning, not personally generating every drawing.

Consider a small accessibility study. A beginner might develop a route through a building or examine an existing proposal with a supervisor, explaining how requirements, dimensions, level changes, and actual use relate. Then they compare alternatives, including generated ones. Which ones rely on unverified assumptions? If several satisfy the requirements, how do they differ in convenience, coordination, or cost? Why recommend one over another? The assignment matters, but so does the exchange around it.

Duo Dickinson’s recent Common Edge essay “What I Learn From Teaching Architecture” emphasizes listening and dialogue rather than one-way instruction. That helps explain what mattered in my early experience: my project architect wasn’t just correcting drawings—he was making room for questions.

AI can also explain corrections and ask follow-up questions. The distinction is not an exclusive human ability to teach. It is accountable oversight: a supervisor must assess whether this beginner can apply an explanation to the actual project and take responsibility for the work under their supervision. A useful answer does not establish that the person receiving it knows how to use it.

That review must distinguish a technical requirement from a project-specific tradeoff or a supervisor’s preference. These are not interchangeable lessons. Experienced professionals should make their reasoning open to questions, not simply demand agreement. Otherwise, anticipating redlines becomes an exercise in pleasing the reviewer rather than understanding the work.

Assignments need a life beyond submission, too. Someone who studies a material should encounter its effects on detailing, installation, maintenance, or use. Someone who develops a layout should see the coordination issues it creates. Where direct participation is impractical, reviewing what happened afterward can still connect decisions to consequences.

The next assignment should reveal whether that experience carries forward. Can the beginner recognize a similar problem under different conditions, explain why the previous solution no longer fits, and know when to ask for help? As that capability develops, they need opportunities to make bounded decisions under supervision, not simply more work to produce. Those opportunities should be deliberate rather than depend on which senior person happens to take an interest.

None of this is an entirely new idea. Good supervision has always required questions, explanation, and progressively greater responsibility. What changes now is the structure supporting them. Production assignments provided occasions for learning and a reason to employ beginners while they developed. If automation compresses those assignments, firms cannot assume the education embedded in them will continue on its own.

Preserving that education is an organizational responsibility. Leadership must provide paid time, access to meaningful work, and realistic supervisory workloads. The format can adapt under project pressure: a brief discussion of a consequential correction may be more useful than a formal mentoring session disconnected from the work. But if explanation always disappears when deadlines tighten, the firm sacrifices the development of people on whose understanding future work will depend. An assigned mentor cannot compensate for that absence of support.

Preserving those learning opportunities also means continuing to hire beginners. A firm may decide it can rely on experienced people trained elsewhere, but the profession cannot collectively depend on someone else to develop its future staff.

Schools provide foundations; judgment continues to mature through encounters with clients, consultants, construction, and the consequences of decisions. If firms stop employing beginners, they remove not only entry-level jobs but the places where that development occurs.

The stakes extend beyond staffing. Faster production does not by itself serve the public. People preparing and reviewing proposals must understand their consequences, recognize what remains unresolved, and know when their expertise is insufficient.

Beginners don’t need to reproduce my stacks of drawings to learn what I learned. They need work that stimulates questions, people who help them investigate those questions, and opportunities to apply the answers under changing conditions. My project architect wanted me to anticipate his redlines. Over time, that meant learning to see a problem before someone marked it for me. If AI produces the drawing, beginners still need a way to learn what deserves a redline—and why.

Featured image courtesy of Lightroom Studio LLC.