What benchmarks matter for AI-generated code/assistance beyond accuracy (e.g., trust, explainability, adoption stickiness)?

I would want to measure AI-powered coding tools in these ways:Performance: time to first useful output, time to respond as context growsTask completion: ability to finish...