The Day ChatGPT Changed Full-Stack Engineering

In late 2022 the cost of writing software dropped overnight. Not gradually, not for early adopters — for everyone, at once. A general- purpose model that could read intent and emit working code turned the single most time-consuming part of engineering, translation, into a conversation. Every assumption about how long a feature takes was suddenly wrong.

From Stack Overflow to a collaborator

The old loop was: hit a problem, search, read three threads, adapt an answer, move on. The new loop collapsed that into a dialogue with a system that already knew the libraries, the idioms, and the edge cases. The friction that defined day-to-day engineering — looking things up, boilerplate, glue code — mostly evaporated. What remained was the part that was always the actual job: deciding what to build and how it should be shaped.

The bottleneck moved to judgment

When typing is cheap, the constraint becomes knowing what to type. The engineers who got faster were not the ones who prompted best; they were the ones who already held a clear architecture in their head and could direct the model toward it. The model amplifies clarity — and, just as ruthlessly, amplifies confusion. Point it at a vague spec and it generates vague software very quickly.

What it meant for MERN engineers

Full-stack developers were positioned perfectly. The MERN surface — typed contracts from MongoDB to React — is exactly the kind of structured, well-understood domain a model navigates well. The engineers who treated the model as a force multiplier on a sound architecture pulled ahead. The ones who treated it as a replacement for understanding the system shipped faster bugs.

The day code became cheap to write was the day architecture became the entire job. Everything since has been a footnote to that shift.

Everything else on this blog — RAG, orchestration, type-safe outputs — is downstream of that 2022 inflection. It is where the work went next.