Generative artificial intelligence (GenAI) has changed the conditions of L2 writing revision and feedback. There is an increasing body of evidence that GenAI support may enhance the quality of text produced under supported conditions. However, an enhanced GenAI-assisted text does not necessarily indicate that the writer has acquired the linguistic, rhetorical, metacognitive, or self-regulatory ability to write independently. A conceptual-critical article reviews the problem of using assisted performance as evidence of writer development in AI-mediated L2 writing studies. The article draws on L2 writing, written corrective feedback, automated writing evaluation, learner agency and emerging GenAI research to distinguish assisted performance from writer development. The argument is that the unit of evaluation should not be limited to the supported performance of the AI-supported text, but it should be extended to what the writer can do later with less or no assistance. The article develops a distinction among assisted performance, revision uptake, learning and independent transfer and concludes with a research perspective in which supported performance, delayed no-AI writing and transfer are considered as different types of evidence. The main claim is that AI-writing studies should consider not only whether artificial intelligence can help learners produce better texts, but also whether its use can lead to the development of writers who can perform comparable writing activities without support.