What Do Developers Want From AI?

Alexander Polomodov and Nikolay Bushkov review What Do Developers Want From AI? and ask why companies rush to add AI across the SDLC without asking engineers what help they need. Nikolay explains how T-Bank’s R&D center studies engineering productivity and demand for new capabilities.

The paper shifts attention from impressive demos to concrete developer jobs and pain points. The discussion combines interviews, quarterly surveys, product logs, and behavioral metrics. This triangulation separates perceived usefulness from actual use and makes the link between Developer Experience and productivity more defensible.

AI capabilities fall into three levels: assisting existing work, extending human ability, and delegating a task. Code review, suggestions, and vulnerability analysis show why each use case needs its own judgment. A model may speed up review while producing weaker code and making the developer more confident that the result is secure.

Technical debt, missing documentation, and tool switching are major obstacles. A retrieval assistant makes internal knowledge easier to use, while unanswered queries reveal what to document next. Adoption works as a product loop: find a pain point, test a hypothesis, enable it on the platform, and measure the outcome.