Microsoft announced that they were banning their developers mostly from the Experience & Devices division from using Anthropic’s Claude code.Mutiple reports from The Verge and Tech radar points to June 30th ,2026 as deadline for ending most access.
According to Rajesh Jha, Microsoft’s executive vice president of experience and devices stated that Claude code was a learning experience that had ran its course The move also aligns with Microsoft’s broader strategy of consolidating developer workflows around products it can shape directly through GitHub, its cloud infrastructure and its own security and governance requirements. However, there is much more than the PR friendly stated explanations
The story began in December last year, when Microsoft introduced Claude Code to its developers to explore how AI-assisted coding could fit within the organization. Access was offered openly, with no specific requirements. At the time, the goal was to let developers experiment and generate ideas as AI-assisted development was still emerging. The most widely cited explanation is the tool became so good that most engineers preferred it over the inhouse option(Copilot CL).Unlike other applications AI uses token generation for computing. Agentic coding tools can generate heavy token usage, and reports suggest that internal spending on Claude code rose significantly with increased usage and therefore they decided to scale it down as part of cutting cost at the end of financial year.
This is not a surprising phenomenon as we have recently seen a pattern of companies burning down their budgets on AI through the token-pricing model which is proving to be expensive. Not long Ago Uber ended up using its entire annual AI budget in 4 months. In a larger scale this move shows how AI coding is entering a new face. After early experimentation that was mainly focused on model qualities and user enthusiasm, now organizations are getting over the ‘vibe check’ to start considering serious ramifications such as cost predictability, platform control, security review
