The blank-check AI coding era is dead. Here’s what comes next.
Microsoft spent the first phase of the AI coding boom pushing its engineers to use the tools. Now it wants The post The blank-check AI coding era is dead. Here’s what comes next. appeared first on The New Stack .
Microsoft has shifted its approach to AI-powered coding tools after the initial enthusiasm of the boom subsided. The company now seeks to determine if the tokens generated by these tools are truly delivering meaningful results. To manage the expense, Microsoft has introduced AI token budgets for its divisions, similar to how it controls other expensive computing resources.
Engineers are now required to use OpenAI's GPT-5.6 Sol as the default model in GitHub Copilot, with executives emphasizing the importance of maximizing outcomes for customers and the business rather than minimizing token usage. The internal memo warns that excessive token consumption is not a priority and encourages employees to focus on impact per token.
Microsoft's decision to make GPT-5.6 Sol the default model complicates the idea that the move is solely cost-focused. Despite being the most expensive model in OpenAI's GPT-5.6 family, it costs less than some previous models. Employees now need to consider the value of each model for their specific tasks, balancing the extra capabilities against the price.
To streamline AI use, GitHub has implemented automatic model selection, routing requests to the most suitable model based on factors like task type and user subscription. This feature helps avoid unnecessary costs and improves token efficiency without compromising quality.
Microsoft has already begun streamlining its internal coding stack, eliminating most Claude Code licenses in its Experiences and Devices division and instructing engineers to switch to GitHub Copilot CLI. However, the company is still investing in coding agents, which can enhance developer output by automating tasks like exploring repositories, building implementation plans, running commands, executing tests, and revising work.
Despite Microsoft's research showing a 24% increase in merged pull requests among engineers using coding agents, it remains unclear whether this leads to better products, fewer bugs, improved security, saved developer time, or greater customer value. As coding agents evolve and become more capable, the cost of usage may outpace the measurement of productivity benefits.
Written by urgent.news from The New Stack's reporting — not their text. Machine-written; read the original for the full account.
