AI Coding Agents

Coding agents split into two shapes and the listings below mix them freely. One sits in your editor and completes what you are already typing. The other takes a task, reads the repository, makes a branch and comes back with a diff. They fail differently: the first wastes a keystroke, the second can spend twenty minutes confidently rewriting the wrong module. Whichever you try, judge it on a repository you know well rather than a demo, and look for how it behaves when the tests fail — that is where the difference between the tools is largest.