Rutube Tech CTO: AI Coding Can Raise the Cost of Software

Using artificial intelligence to build software can make the final product more expensive, according to Dmitry Egorov, chief technology officer of Rutube Tech. His comments were relayed by the press service of Gazprom-Media Holding's Digital Assets division.

Egorov says model-driven development can speed up the first stage of a project. But that gain is offset by the time and money needed to find errors and to maintain the finished product.

Accumulating errors are the main risk

The biggest danger, in his view, is accumulated mistakes that a model is unlikely to fix on its own. A team can quickly get an initial version whose internals are hard for a human to understand, so later changes are also made mostly by AI. Errors and inefficient solutions then build up gradually.

Egorov added that using two separate models, one to generate code and another to check it, does not guarantee a given level of quality. Evaluating the work of AI agents properly requires a specialist with expertise in the relevant field, and AI will not replace such professionals.

He sees AI as useful for small teams that need to generate source code and test business ideas. For complex, long-lived systems, however, the reliability and stability of its output must be confirmed first.

Compute and security add to the bill

Shorter development time does not mean cheaper production, Egorov said. AI use significantly increases spending on computing power. Project budgets should also cover not only the running of the models but also cybersecurity verification of the software. Trying to save money with AI-generated code could end up raising verification and maintenance costs.

He also noted that AI models will only increase the workload of the specialists who review the output. In his opinion, models are better used for detecting vulnerabilities, handling cyber incidents and configuring alerting systems.