Learning to Code in the Era of Large Language Models

Mark Seemann answers a reader’s extensive letter about learning programming while large language models (LLMs) are widely available.

A Reader’s Experience with AI‑Assisted Development

The correspondent writes, “I’m trying to understand how people who deeply understand software think about learning and competence in the age of AI.”

He adds, “I’m approaching it almost as a historian would: asking people directly how they make sense of a technological transition while actually living through it.”

About a year ago he became fascinated by AI‑assisted programming despite lacking a formal computer‑science degree.

Using LLMs he assembled a sizable TypeScript/JavaScript system that integrates APIs, PostgreSQL, LLM pipelines, research automation, and multi‑model workflows.

He describes the early phase as “almost magical,” noting that AI seemed to collapse the distance between an idea and a working implementation.

When he began converting the prototype into a production‑grade product, he encountered a cascade of errors that each required another model query to resolve.

After months of refactoring he realized the system may have exceeded his personal level of comprehension.

He observes that the gap between the system’s functionality and his understanding is invisible when everything works, but becomes starkly apparent during failures.

He confesses, “Sometimes I genuinely don’t know what to do next without asking another model,” and wonders whether he has built a functional product or merely the appearance of one.

He states, “I’m not anti‑AI at all. I’m fascinated by these systems and want to work with them professionally.

But I’m unsure what the right relationship with them should be.

The Author’s Perspective on LLMs and the Workforce

Seemann discloses his own ambivalence, saying he leans toward disliking AI while acknowledging its unstoppable momentum.

He notes that LLMs impress him at times and frustrate him at others, especially when they challenge the relevance of his three‑decade‑long expertise.

He writes, “When it’s bad, it can be frustrating, but then at least I can absorb an ember of warmth in the illusion that what I’ve spent more than thirty years learning is still relevant.”

He adds, “When it’s at its best, I sometimes think: Where do I sign up for the Butlerian jihad?” indicating a personal tension between admiration and alarm.

Seemann frames his view through an economic lens, noting his age and accumulated success give him a safety net that many knowledge workers lack.

He cautions that programming may be the first white‑collar field where verification is easier, potentially leading to earlier job displacement than in other sectors.

He references historical technological shifts, comparing possible AI‑driven unemployment to the impact of the stocking frame, steam engine, internal combustion engine, and computers.

He acknowledges the classic argument that technology creates new, unforeseen jobs, but counters that those new roles rarely match the displaced workers’ skills.

He cites, “Coal miners didn’t just become programmers overnight,” to illustrate the mismatch between old and new occupations.

He stresses that mass unemployment of 30‑40 % would have a profound economic impact, a scenario he finds difficult to imagine a stable society surviving.

Implications for Aspiring Programmers

The exchange highlights a growing dilemma: LLMs can accelerate prototype creation, yet may also produce systems that outpace a developer’s depth of knowledge.

For newcomers, the letter suggests the importance of balancing rapid AI assistance with deliberate practice of underlying concepts.

Seemann’s reflections imply that reliance on models without solid fundamentals could hinder long‑term employability as the market filters out shallow expertise.

Conversely, his acknowledgment of AI’s productivity gains indicates that mastering prompt engineering and model interaction may become a valuable skill set.

Readers should watch how educational programs and industry standards evolve to address the need for both AI‑augmented efficiency and rigorous software understanding.

Why This Matters