Paul Ford reports that in November, Claude Code improved noticeably enough that he was able to complete side projects that had been untouched for a decade or more.
Paul Ford reports that AI coding tools shifted from being halting and clumsy to being able to run for about an hour and produce whole designed websites and apps that are flawed but credible.
Paul Ford reports that large language models can handle expensive tasks such as migrating data from an old platform to a modern one.
LLM-driving skills are becoming more important than detailed knowledge of specific software platforms.
It is currently unclear whether LLM-driven development will increase recognition and importance of Expert Generalists.
It is currently unclear whether LLMs will reduce organizational silos or instead enable teams to work around silos by generating large amounts of code.
Typing Tradeoffs Shift With Agent Labor
After more than 25 years as a programmer, the author reports starting to prefer type hints or strong typing more than before.
The author reports that explicitly defining types becomes significantly more attractive when a coding agent performs the typing work.
The author reports that a prior reason for avoiding type hints or strong typing was that they reduced iteration speed, especially in REPL-centric workflows.