Member-only story
I have not hand-written much code lately.
That does not mean I stopped engineering. I spend more time reviewing code that an AI wrote, deciding how a system should behave, and iterating on agents that use tools, memory, prompts, APIs, context windows, budgets, retry logic, and verification. The code still has to be good. It just arrives faster than it used to.
The bottleneck moved. I can ask for a feature, get a credible first implementation, and then spend the next hour on the work that decides whether it belongs in the product at all. What should the agent be allowed to call? What customer state is it allowed to change? Which source wins when two tools disagree? When should it stop instead of spending another dollar on tokens? What counts as proof that the feature works?
That is the part of software engineering I care about going into 2027 and 2028.
I am not claiming that engineers vanish or that every product manager suddenly becomes a systems architect. Nobody has clean evidence for a net employment forecast that specific. Lower build cost can mean fewer people for one fixed project, more projects attempted, or both. Those are different outcomes.