A joint study by T1 and Tproger reveals that 42% of developers struggle to understand code generated by AI. This issue is cited more frequently than AI errors themselves. Additionally, 39% of respondents report that reviewing AI-generated code takes more time than expected, while 36% note that AI tools lack necessary context regarding existing codebases and team standards.

The complexity of AI-generated code is particularly high when using AI agents for full tasks, with 51% of users reporting comprehension difficulties compared to 31% of those using AI only for suggestions. Conversely, 59% of users relying on AI for hints encounter errors and hallucinations, whereas only 17% of those delegating full tasks report such issues.

Organizational challenges remain significant, as 43% of developers state their companies lack clear rules for AI usage, and only 10% have mandatory result verification policies. Despite these hurdles, 73% of developers use AI tools daily, and 78% report increased speed in completing tasks. However, only 8% received formal corporate training on AI tools.

T1 plans to transition from ad-hoc AI usage to a structured IT production model that integrates AI across all development stages while maintaining security standards. Sergey Polinenko, product leader of the Sfera platform, stated that the company is currently working on making AI-driven coding a trusted process for business needs.