"AI coding tools are getting better fast. If you don't work in code, it can be hard to notice how much things are changing, but GPT-5 and Gemini 2.5 have made a whole new set of developer tricks possible to automate, and last week Sonnet 2.4 did it again. At the same time, other skills are progressing more slowly. If you are using AI to write emails, you're probably getting the same value out of it you did a year ago."
"The difference in progress is simpler than it seems. Coding apps are benefitting from billions of easily measurable tests, which can train them to produce workable code. This is reinforcement learning (RL), arguably the biggest driver of AI progress over the past six months and getting more intricate all the time. You can do reinforcement learning with human graders, but it works best if there's a clear pass-fail metric, so you can repeat it billions of times without having to stop for human input."
AI coding tools are advancing rapidly due to reinforcement learning methods that can leverage billions of automated, measurable tests. Models like GPT-5, Gemini 2.5, and Sonnet 2.4 enable automation of many developer tasks. Reinforcement learning benefits when tasks have clear pass-fail criteria, allowing large-scale iterative training with minimal human input. Capabilities that are RL-friendly—bug-fixing and competitive math—are improving quickly. Skills without easy automated grading, such as writing emails or multi-role chatbots, are progressing far more slowly. The resulting reinforcement gap is creating a growing divergence in what AI systems can and cannot do. Software development aligns well with RL because testing is already integral.
Read at TechCrunch
Unable to calculate read time
Collection
[
|
...
]