From “Replacing Programmers” to “Managing Costs”: AI Coding Enters a New Phase

 In recent years, AI programming tools have garnered significant attention as a force that will dramatically change development efficiency. With the emergence of tools like GitHub Copilot and Claude Code, developers can now generate and debug code with simple instructions. This has led to increased discussion about whether AI will take away programmers’ jobs.

However, a shift in this trend is beginning to emerge. GitHub has announced a move to shift Copilot’s pricing model from a traditional subscription-based model to a usage-based (token-based) model. Anthropic has also imposed restrictions on the use of high-performance models in Claude Code, sometimes requiring additional fees.

The underlying reason is the high computational cost of AI.

Large-scale language models consume “tokens” for input, output, and inference. Furthermore, automated tools like AI agents perform inference and tool calls multiple times internally, potentially consuming large amounts of tokens without the user’s knowledge.

This trend is even more pronounced in enterprise environments. When AI agents automatically execute multiple tasks, a massive amount of tokens can be used in a short time, potentially leading to a sharp increase in costs. Therefore, some companies are beginning to set budget caps on AI usage.

For example, if the annual cost of one engineer is approximately $250,000, that comes out to about $20,000 per month. If an AI tool can significantly improve productivity for just a few hundred dollars, it’s extremely effective. However, if a team’s AI usage costs tens of thousands of dollars per month while the results are limited, AI could become a new expense rather than a cost-saving measure.

In other words, the important thing is not “whether to use AI,” but “how to use it”.

Optimizing how AI is used is required, including proper prompt design, agent usage management, and reducing unnecessary model calls.

AI is certainly a powerful tool for increasing development efficiency. However, its value depends on whether the results it produces outweigh the costs. Going forward, cost management, as well as efficiency, will likely be crucial themes in AI utilization.