Agent-based development diffusion… companies are eager to build multi-model AI and governance systems

In the front lines of software development, “agent-based development” is rapidly spreading, and the ways companies incorporate AI are also undergoing significant changes. Diagnostics indicate that, compared to relying on specific vendors, building a “multi-model AI” system that operates multiple AI models and agents simultaneously is becoming a core focus.

Mikhail Vink, Vice President of Business Development at JetBrains s.r.o., pointed out in an interview at Google Cloud Next that recent changes in the development environment are far exceeding market expectations. JetBrains is a global integrated development environment (IDE) company used by 15 million developers. Vink stated, “This month it’s Anthropic, next month it’s Gemini, and new AI features are emerging one after another,” and emphasized that developers need to be able to flexibly utilize multiple models to achieve the best market results.

He specifically explained that the real challenge of “agent-based development” lies not so much in code generation itself, but in the infrastructure built afterward. For companies to apply AI to actual business operations, they must control multiple agents, connect data, context, and memory layers, and even organically integrate external tools and pipelines. This means that beyond simple experimental levels, enterprise-level AI environments require complex structures that far exceed expectations.

Code generation has become simpler, but operation and control are more important

Vink emphasized that for agents to operate normally, “context” and “actual data” are essential. To this end, he believes it is necessary to simultaneously establish connections with the Model Context Protocol (MCP) server, transmit structured data, and set up sustainable development environments. He explained that without these fundamentals, AI-generated results may diverge from real-world work.

JetBrains stated that in response to this trend, they are building a governance platform to track costs, manage model access permissions, and analyze how developers actually adopt AI suggestions. Their judgment is that, from an enterprise perspective, only by clearly understanding AI usage, efficiency, and control levels can large-scale adoption be achieved. This means that as AI applications expand, “governance” and “visibility” are as important as performance competition.

The role of developers is shifting from mere coders to “orchestrators”

This change is also transforming the role of developers themselves. No longer just stopping at directly writing code, the role of coordinating multiple AI agents and models as “orchestrators” is becoming increasingly important. At the same time, requirements for quality assurance, security validation, and understanding algorithms have reached higher levels.

Vink regards “critical thinking” as the most important factor in quality management. He believes that instead of directly approving results provided by AI, one should deeply understand how the system works and personally verify the generated algorithms. This indicates that even if AI improves development productivity, ultimate responsibility and judgment still rest with human developers.

Ultimately, in the era of agent-based development, competitiveness does not solely depend on using the latest models. Systematic enterprises that can flexibly combine multiple AI models, securely control them, and manage output quality are more likely to grasp the initiative in software development in the future.

TP AI Notes: This article is summarized using the TokenPost.ai basic language model. The main content may be omitted or inconsistent with facts.