Look at our scoreboard and one category stands apart. AI code editors are the only group where every major player clears an 8.4 overall — Cursor at 9.0, Windsurf at 8.5, Copilot at 8.4. No other category is close to that density of quality. It is worth asking why.
Feedback loops you can measure
Code either runs or it does not. That brutal feedback loop means editor teams know within hours whether a model change helped, and users feel improvements immediately. Compare that to writing tools, where "better" is a matter of taste and the ceiling is set by how good the underlying model feels in prose.
The buyers are the builders
Developers building AI tools are their own first customers. The people shipping Cursor use Cursor to ship Cursor. That dogfooding loop compounds: every rough edge gets felt by the team that can fix it the same day.
Competition without a default winner
GitHub could have owned this category by default. Instead, Copilot's big-company release cadence left room for Cursor and Windsurf to out-ship it, and now three serious players push each other weekly. Categories with a lazy incumbent — or no incumbent — do not get this effect.
What it means for everyone else
Video generation is starting to show the same pattern: measurable output quality, fierce competition, fast iteration. Writing tools are not — which is why the category's scores cluster in the 7s. When you evaluate any AI tool, ask the editor-category questions: is there a hard feedback loop? Do the builders use it themselves? Is anyone pushing them?
The categories that can answer yes are where the 9s will come from.