What People Are Actually Using AI For Right Now

December 8, 2025

25 min episode · 2 min read

Episode

25 min

Read time

2 min

Topics

Artificial Intelligence

AI-Generated Summary

Published Dec 21, 2025

Key Takeaways

✓Reasoning Model Adoption: Reasoning models grew from negligible to over 50% of token consumption in 2025, representing a complete paradigm shift in how users engage with AI systems.
✓Programming Dominance: Coding applications expanded from 11% to over 50% of usage, with average prompt length growing 4x as users submit complex codebases rather than simple requests.
✓Model Selection Strategy: Teams use closed models for high-value workloads and open-source models for high-volume tasks, with Chinese open-source models capturing up to 30% market share weekly.
✓User Retention Patterns: Models that first solve painful workloads create permanent lock-in, with early Claude and Gemini cohorts retaining 40-50% of users after six months despite newer alternatives.

What It Covers

Analysis of 100 trillion tokens reveals real AI usage patterns: reasoning models jumped from negligible to 50% usage, programming dominates at 50%, role-playing leads open-source consumption.

Key Questions Answered

•Reasoning Model Adoption: Reasoning models grew from negligible to over 50% of token consumption in 2025, representing a complete paradigm shift in how users engage with AI systems.
•Programming Dominance: Coding applications expanded from 11% to over 50% of usage, with average prompt length growing 4x as users submit complex codebases rather than simple requests.
•Model Selection Strategy: Teams use closed models for high-value workloads and open-source models for high-volume tasks, with Chinese open-source models capturing up to 30% market share weekly.
•User Retention Patterns: Models that first solve painful workloads create permanent lock-in, with early Claude and Gemini cohorts retaining 40-50% of users after six months despite newer alternatives.

Notable Moment

OpenRouter data shows role-playing and creative writing comprises 52% of open-source model usage, revealing a massive gap between venture capital productivity focus and actual consumer behavior.

Know someone who'd find this useful?