Somebody finally asked CTOs which AI model actually runs their engineering work.
Claude took 68 percent. Gemini, Llama, DeepSeek and Mistral each got zero. It's only 41 CTOs though, so don't read it as market share.
New tool: DeepSeek Harness.
Everything is a plugin.
When people whining about DeepSeek V4.1 Flash not being good enough, I think it is a skill issue.
Created with DeepSeek v4.1 today:
5-hour limits? Where we're going, we don't need limits.
GB300 = Machine Gun Kelly for tokens. Rat-tat-tat on your desk. No UI theater. No spinners. Just raw GPU violence with DeepSeek V4.1 Flash. Local AI for enterprise just went supersonic
You have reached the end of the archive
All of deepseek