When AI Can Write 90% of Code, Where Does a Programmer’s Value Lie?

A colleague showed me a piece of code generated by Cursor and asked why it wouldn’t run. The code itself was well‑structured, with clear logic, proper naming, and thorough comments, but it called an internal service that had been deprecated three months earlier. The AI had no awareness of this change, producing a perfectly written yet unusable snippet.

This illustrates a subtle problem: AI can write "90‑point" code in an ideal, fully documented world, but the real work environment is riddled with legacy systems, temporary hacks, commented‑out logic, and undocumented "do‑not‑touch" sections that exist only in collective memory. AI has never walked through those pitfalls.

The truly hard part of a programmer’s job is not implementing algorithms like quicksort or an LRU cache—tasks AI already handles well—but making judgments when faced with vague requirements, half‑broken systems, conflicting stakeholder demands, and tight deadlines. These judgments involve weighing technical debt, team capabilities, business intent, and intuition about a system’s future direction.

AI can assist by writing implementations, yet it cannot replace the decision‑making process that defines a developer’s role. That judgment is the reason programmers remain indispensable.

Some coding work is purely repetitive, such as writing CRUD operations many times. Letting AI handle this "mechanical" labor frees developers to focus on the nuanced decisions that AI cannot make.

Many engineers who use AI effectively share a common trait: they know exactly what they are doing. They generate an initial draft with AI, then read every line, ask the AI why it chose a particular approach, and evaluate whether the reasoning holds. They use AI to accelerate thinking, not to replace it.

This approach actually demands stronger fundamentals, because developers must be able to audit the "90‑point" code and locate the missing "10 points"—the hidden assumptions, legacy constraints, and contextual knowledge that AI overlooks.

In summary, a programmer’s value no longer resides in the ability to write code alone—the barrier to entry is lowering—but in deep system comprehension, accurate business perception, and the accumulated implicit knowledge that AI cannot acquire.