The Decisions That Remain Irreducibly Human
The conversation about AI in software development tends to collapse into two positions: either AI will replace developers, or AI is just a faster autocomplete. Both positions avoid the more interesting question, which is about the specific category of decisions that cannot be delegated and why.
Some decisions remain irreducibly human not because the technology is insufficient but because the decision requires accountability, context, and judgment that exist outside the codebase. Whether to ship a feature with a known edge case. Whether a performance tradeoff is acceptable given who the users are. Whether the architecture chosen today will constrain the business in ways that are not yet visible. These are not technical questions with correct answers. They are judgment calls that require someone to own the consequences.
The category also includes decisions that require understanding the gap between what was specified and what was meant. AI systems are very good at implementing specifications. They are not good at recognizing when a specification is subtly wrong in ways that the author did not intend — when the requirement, if implemented literally, would produce behavior that no one actually wanted. A developer who has been on a product for two years brings institutional knowledge to that gap. A generation model does not have it, cannot have it, and will produce exactly what was asked for whether or not that is the right thing.
This reframes what seniority means in an AI-assisted team. The most valuable engineers are not the ones who can generate the most code — that is now cheap. They are the ones who can identify the decision points that matter, ask the questions that prevent costly misdirection, and maintain the architectural vision that keeps a system coherent over time.
The craft of software development is not disappearing. It is concentrating. The work that remains is harder and more consequential, not easier, and it requires the kind of judgment that only comes from sustained engagement with the problem and the domain. That is not a threat to the profession. It is a clarification of what the profession is actually for.