Engineering write-ups, case studies, and the occasional opinion. Every article here was drafted in the backstage and published by a human.
The Claude app, Claude Code, and Claude Design run on the same family of models; what differs is the harness around them. Using the job of organizing 40 interview transcripts, we show how each harness decides what the AI can reach, what it can do, and where the results go.
The same AI model can behave very differently from one product to the next, and the difference usually sits outside the model. We read the source of DeepSeek Harness and Google's Harness², and use them, along with our own Beluga, to explain what a harness does and how to judge one.