Developers are losing productivity due to "tokenization."

Research shows that while developers generate more code thanks to AI, their actual productivity decreases due to increased rework.

In the programming world, debates are heating up about how developer productivity is measured in the era of active artificial intelligence adoption. A new study signals that while software creators may produce more code, they are turning to rework much more frequently, which calls into question the effectiveness of new tools.

Productivity analytics show that the use of AI agents, such as Claude Code and Codex, does indeed increase the volume of generated code. However, as Alex Tsirceya, CEO of Waydev, notes, after corrections, efficiency drops to 10 — 30%. More than 80% of the code created by AI is ultimately accepted by developers, but the need for corrections reduces their productivity.

Under pressure from fierce market competition, Waydev is revising its strategy by introducing new tools for metadata analysis. This will allow companies to better understand the quality and cost of generated code, which has already attracted the attention of major players like Atlassian, which recently acquired the startup DX for $1 billion to increase transparency in its AI investments.

Data from GitClear and Faros AI confirm the critical dependence on AI tools: with nearly a 10-fold increase in productivity, the volume of removed code increases by 861%. This threatens the key ambitions of developers seeking to optimize their workflows.