OpenAI reveals coding agents reshaping AI research with early data on agent usage, experiment velocity and task complexity driving acceleration.

According to @gdb, Codex hackathons demonstrate remarkable builder energy, highlighting groundbreaking achievements by top teams among over 200 participants. Innovations ranged from C++ firmware for brainwave readers to orchestrating fleets of coding agents. Participants showcased the potential of AI technologies, emphasizing advancements in automation and real-time applications.

According to @DeepLearningAI, its year-end The Batch edition frames 2025 as the year reasoning models changed everything and explores how AI learned to think before it speaks (source: DeepLearning.AI). The edition highlights three themes: China turning chip restrictions into innovation fuel, coding agents becoming true development partners, and U.S. economic growth driven by infrastructure spending (source: DeepLearning.AI). It also features a holiday message from Andrew Ng to the AI community (source: DeepLearning.AI). For traders, these covered themes map directly to compute supply chains, software automation, and capex cycles that shape positioning in AI equities and AI-linked crypto narratives (source: DeepLearning.AI).