Q and A with the Experts: What is “Claudish”?

AI coding agents use a distinctive style of English. The terms exist in software engineering and other technical areas, but how frequently Claude uses them, the way it combines them and the compressed style it uses are distinctive. Professor Yuntian Deng, from the Cheriton School of Computer Science at the University of Waterloo created a translator to help readers better understand Claude’s information-dense outputs. Professor Deng answers some questions about Claudish and his new translator, which has received over a million views on X.   

What is “Claudish”?  
People have nicknamed the style of English Claude Code and other coding agents tend to use “Claudish.” It often relies on compact technical vocabulary and expressions like "gated on" or "load-bearing", words from existing technical fields, and uses them in very concise statements describing fairly complicated relationships. The result is information-dense output that’s surprisingly difficult for a human reader who hasn’t spent much time interacting with coding agents.   

Why is it important for people to become familiar with the language?  
There are practical benefits. Once you become familiar with the style, you can understand an agent's compressed technical output immediately rather than repeatedly asking it to explain itself in simpler language.   

The broader phenomenon is more interesting: language generated by AI systems may actually start influencing the way humans communicate with one another. For decades, we tried to make computers understand the way humans naturally communicate. Now people who spend hours every day working with AI agents are beginning to learn the agents' preferred vocabulary and communication patterns themselves.  

Why did you decide to create a translator for Claudish?  
I kept noticing Claude using the same kinds of expressions repeatedly and then I started noticing myself and other people in tech using some of them too. When I looked more carefully, I realized that many words already existed in different technical areas; they were just much less familiar to me. That made the phenomenon even more interesting: Claude can draw on precise terminology from many different communities in a way that no individual necessarily could. I thought building a translator would make the phenomenon much easier to see. The demo also includes a dictionary of some of the expressions I've been noticing. The translator is built using ProgramAsWeights, a research project from my group at Waterloo that studies how natural-language specifications can be compiled into small local neural functions.  

Are we going to hear Claudish word use in casual language?  
We already are, at least within parts of the tech community, and I've done it myself. For example, I now catch myself describing things as "[something]-gated" or "byte-identical". After spending so much time interacting with coding agents, these kinds of expressions start to feel increasingly natural. As people spend more of their working lives communicating with AI agents, how much will the agents begin to shape human language as well? Or, in Claudish, how "agent-shaped" will human language become?