Autonomous AI agents are capable of independently taking on developer tasks, changing code, running tests, and submitting results for review

The Yandex B2B Tech team has added autonomous AI agents for code development and verification to the SourceCraft platform. They can act as independent team members: receive tasks, execute them, run tests, and submit results to developers for review.

An agent can be assigned directly in a GitLab task. Additionally, it can be interacted with via the SourceCraft web interface, messengers, VS Code, and the command line. If additional data or approval is required to complete the assignment, the agent will independently contact colleagues.

For example, upon detecting an error, a developer can assign an AI agent as the task executor. It will prepare the environment, make changes to the code, conduct testing, and attach the result to the original task. After review by a specialist, the agent will be able to take into account comments and refine the solution.

According to McKinsey, companies that have implemented autonomous AI agents in development were more than twice as likely to report productivity growth exceeding 20% compared to organizations using AI only as a separate tool. On average, agents reduce task completion time by 11.2% and rework volume by 6.8%.

Yandex B2B Tech notes that AI is gradually being used not for individual operations, but for automating entire development chains. At Yandex itself, 73% of developers regularly use AI, and generative models are involved in more than half of the code created.

In the future, the company plans to combine access to SourceCraft, Alice AI Pro for Business, and other AI services into a single subscription with the ability to use Yandex AI Studio models.