Anthropic doubles Claude Code limits as user growth outpaces compute plans
The AI firm abandons long-term roadmaps in favour of rapid experimentation, citing model intelligence over token efficiency as the primary driver for its product philosophy.
Anthropic has doubled usage limits for Claude Code on its Pro and Max subscription plans, a move designed to alleviate compute constraints following an 80-fold surge in user growth. The announcement was made at the company’s second annual Code with Claude conference in San Francisco, where CEO Dario Amodei noted that the rapid expansion far exceeded internal projections of 10x annual growth. This discrepancy has strained infrastructure, prompting the adjustment to restore service stability for developers utilising the agentic software tool.
Cat Wu, Anthropic’s head of product for Claude Code, stated that the company does not maintain a long-term roadmap for the product. Wu explained that such fixed plans would likely become obsolete given the exponential pace of model improvements. Instead, the firm employs a "lean harness" approach, prioritising raw model intelligence over token efficiency. This strategy allows for rapid, experimental feature development, with the team shipping updates in weekly cycles based on real-time developer signals rather than adhering to a rigid annual plan.
The company also addressed user frustration regarding high token consumption, identifying two primary causes for the compute crunch. Wu noted that broken session caches significantly increase the cost of resuming conversations, while some users were inadvertently running excessive subagents through plugins. To improve transparency, Anthropic is introducing notifications to alert users when caches are broken or when high subagent usage is detected, helping developers better manage their consumption patterns.
While the command line interface remains the centre of gravity for power users due to its speed and feature depth, Wu observed a gradual migration towards desktop applications. Developers are increasingly adopting graphical interfaces to manage complex, multi-agent workflows, which often involve monitoring multiple terminal tabs and notifications. The company is also exploring "anticipatory" features, where Claude Code could proactively monitor feedback channels such as Slack and GitHub to identify bugs or feature requests without explicit user setup.
Anthropic’s strategy contrasts with competitors who are embedding more structured semantic data into their tools. Wu indicated that the company removes opinionated scaffolds as models improve, believing that domain-specific structures can become counterproductive. The firm remains focused on maintaining a minimal viable set of core tools, allowing developers to install custom plugins only if they demonstrably improve intelligence or efficiency. This approach aims to provide maximum utility per token while adapting to the evolving needs of users ranging from individual coders to large enterprises.
