1. AI coding tools are rapidly advancing due to reinforcement learning (RL), which uses billions of tests to improve automated code generation. Notable advancements include GPT-5, Gemini 2.5, and Sonnet 2.4.

2. Progress in other areas like email writing and chatbots is less rapid. This discrepancy arises because RL works best with clear pass/fail metrics, which are easier to implement in coding tasks than subjective ones like writing.

3. The gap between skills that can be automated using reinforcement learning (RL-friendly) and those that cannot (like writing) is growing. This "reinforcement gap" will impact both startups and the economy as a whole, with automatable processes becoming more common in areas such as software development and healthcare services.