Coding agents ship production work, open models tighten the gap, and Copilot updates 59 times.
This was the week the agent conversation stopped being theoretical: production coding agents shipped real work, Microsoft pushed 59 Copilot updates in one cycle, and the open-weight ecosystem answered the frontier with another release. Three fronts, one direction.
The coding-agent race got its own head-to-head this week: GitHub Copilot against Cursor on the same eight tasks in the same repositories. The multi-file agent gap is real โ and so is the price gap that keeps Copilot in most tool belts.
Cursor Pro: 2 months free on annual (code CURSOR2FREE) โ the editor our multi-file tests scored highest. Read the head-to-head before you decide which seat to buy.
Agents stopped being demos this week. The maintenance habit is the new skill.
A reader asked why our coding comparisons use private repos: because public benchmarks are trained toward. Our test repos are ours โ the failures are too.