# Learning to code when AI writes the code

> A veteran programmer wrestles with what it means to learn software development in the age of large language models.
- Title: Learning to code when AI writes the code · Ploeh | Shortify
- Summary: A veteran programmer wrestles with what it means to learn software development in the age of large language models. AI can help you build faster than you can…
- Keywords: mark seemann, llms, programming education, ai-assisted coding, software development, typescript, javascript, career advice, dunning-kruger effect, learning fundamentals, technology, startups
- Source: Ploeh — https://blog.ploeh.dk/2026/09/16/on-learning-programming-in-an-age-of-llms
- Author: Mark Seemann
- Read time: 2 min
- Topics: mark seemann, llms, programming education, ai-assisted coding, software development, typescript, javascript, career advice, dunning-kruger effect, learning fundamentals, technology, startups
## A reader's AI dilemma
A programmer with no formal training used LLMs to build a large TypeScript system. Now turning it into a real product, they're struggling.

> I may have built a system that is above my own level of understanding.
## The illusion collapses
Fixing one AI-generated error creates another. After months of refactoring, the gap between building and understanding became impossible to ignore.
## Building vs owning code
When everything works, not understanding your own system is invisible. When it breaks, you realize you don't know what to do without asking a model.
## A veteran's ambivalence
Mark Seemann hasn't decided on AI but leans toward disliking it. He works with it, experiments, and finds it often impressive but sometimes frustrating.

> It's usually when it impresses me the most that I resent it maximally.
## When AI works too well
When LLMs are bad, at least thirty years of learning still feels relevant. When they're at their best, he thinks about signing up for the Butlerian jihad.
## The mass unemployment question
Programming may be where LLMs take jobs first, since verification is easier than other fields. But 30-40% knowledge worker unemployment would reshape society.
## History's false comfort
New technology creates new jobs, but not for those who lost theirs. Coal miners didn't become programmers overnight.
## What to learn today
If starting over, Seemann would consider carpentry, metalworking, or gunsmithing. Replacement of manual labor seems farther in the future than knowledge work.
## Abstractions all the way down
Developers have always worked on abstractions they didn't fully understand. The rule: know the level directly below yours and the one above.
## Building faster than learning
AI lets people develop faster than they can understand. But this isn't entirely new to software development.
## The decades it takes
Seemann spent years balancing getting things done while learning on the job. Reaching real competency took decades. Do you have that much time today?
## The human bottleneck
LLMs let you ask more directed questions than buying books hoping for helpful material. But human brains can only absorb new knowledge so fast.
## Learning by trial and error
Seemann learned slowly through trial and error, backed by books. He picked up Basic dialects and C# from documentation and examples alone.
## LLMs bullshit, not hallucinate
Seemann deeply distrusts LLM answers. He asks falsifiable questions with verifiable answers, like whether code works or is shorter, not what to learn next.

> LLMs don't hallucinate; they bullshit.
## Key takeaway

AI can help you build faster than you can understand what you've built, creating systems you can't truly own or debug.