AI made programming differently difficult, not easier

Jeremy Osborn argues in Communications of the ACM that the common claim, that AI has made programming easier, misreads what actually changed. The difficulty did not disappear. It moved. Coding assistants now handle the parts that used to fill a working programmer's memory, syntax, boilerplate, the exact shape of an API call, so recall matters less. What they cannot do for you is decide whether the structure is right, whether the pieces fit, or whether the generated code should be trusted at all.

His frame is that programming has become a task shared between a person and a machine, and the human's share is the harder, more explicitly intellectual half. Osborn describes a shift from recall to judgment, from syntax to architecture, and from implementation to orchestration. The programmer moves from being a knowledge vessel, someone who holds the language and the patterns in their head, to an orchestrating agent who directs tools and checks their output. That is not a lighter job. It removes the mechanical friction that once gave beginners a clear ladder to climb, and replaces it with work that rewards experience and taste.

The argument is worth reading alongside the daily experience of using these tools. Anyone who has watched an agent produce plausible code that is subtly wrong knows the new bottleneck is not typing, it is judging. You can read Osborn's piece here.

Why it matters

If you manage or mentor engineers, the skills that now decide who is good have shifted toward architecture and judgment, so hiring and training built around syntax fluency and quick recall will increasingly miss the point.