KAMAZ has divided digitalization into classical automation and AI autonomy. The first operates according to strict rules and macros, where a person controls every step. The second is built on data-driven learning: the agent understands the task itself, selects tools, and adapts to conditions. This was stated by Nikolay Lunyov, Director of Consulting and Integration at the KAMAZ Digital Products Center, at the “Digital Days of Innopolis 2026” forum in Tatarstan.
The company has identified four levels of autonomy. The first is a chat assistant that searches for data and provides summaries, leaving the decision to a human. The second is skills that perform typical expert operations according to regulations. The third is an agent in the process: it is integrated into the work environment, collects information, and prepares drafts. The fourth is autonomous action, where AI makes decisions within defined boundaries, and a human only addresses deviations.
The main priority is generative AI at the early stages of a product's life cycle: the cost of correcting an error grows exponentially, from a conditional unit at the concept stage to 2000X in serial production.
Practice confirms the reliance on artificial intelligence. The AI assistant processed over 229 thousand records and scans of equipment passports, including handwritten ones, and cleaned duplicates for loading into 1C. Norm control accepts a drawing and issues an inspection report for seven groups of requirements of the unified system of design documentation (ESKD) with references to GOST clauses. The agent in Failure Mode and Effects Analysis (DFMEA) reduced approval from 6–10 days to 2–3. Security is ensured by a hybrid infrastructure within the factory perimeter: data and design documentation do not leave the enterprise, and models are retrained on internal terminology.
In addition, employee training is underway – the course “AI for Everyone” is available on the ai.kamaz.tech portal, lasting 1 hour and 12 minutes.