Anthropic Growth Head: Claude Subscription Usage Has Fundamentally Changed
Amol Avasare, Anthropic's Head of Growth, reflected in several public statements that Claude Max was initially designed solely for high-intensity chat use, without features like Claude Code, Cowork, and long-running agents included at that time. He noted that since the release of Claude Code in Max, the launch of Cowork, and the integration of long-term asynchronous agents into daily workflows, the actual usage patterns of users have fundamentally changed, with the overall usage intensity of individual subscription users significantly increasing. The company is only conducting small-scale tests and exploratory plans to find new structural arrangements to ensure sustainable experiences.
Source: Public Information
ABAB AI Insight
Avasare's statements essentially highlight a fact: Anthropic's early subscription pricing model was designed for "heavy chat" use, rather than for a composite workload of "code editing + file system operations + long-term agents." This means that many heavy users are currently running a "quasi-cloud computing/automation platform" workload under a "chat pricing" shell, which is inevitably unsustainable in terms of computing costs and profit models.
From a cost structure perspective, the resource consumption characteristics of Claude Code, Cowork, and long-term agents are completely different from ordinary conversations: longer contexts, more frequent calls, longer task durations, and often tied to specific workflows rather than scattered queries. In this case, if the early unified pricing is maintained, the marginal usage costs will be largely consumed by a "heavy minority of users," leading the subscription product to inevitably face hidden throttling, declining experiences, or be forced to raise prices significantly.
Avasare emphasized that "engagement per subscriber has significantly increased" and that "the current plan was not designed for this," indicating that what Anthropic is doing is a form of "resource re-matching": shifting the pricing logic from being based on "chat sessions" to focusing on "real computing consumption" and "workflow value density." This will drive subscriptions from a simple "monthly fee for a certain amount" to a more complex tiered structure—different features and intensities of use will be split into different price tiers and quota pools.
In the longer term, these statements clearly expose the migration path of the entire industry: the first couple of years were the era of "general conversational AI subscriptions," and as capabilities like Claude Code, Cowork, and agents mature, the focus of subscriptions has quietly shifted to the "AI productivity stack." In this new structure, those who can align pricing models, computing scheduling, and real workflows will have a better chance of building stable unit economics and stronger sticky moats; users will gradually transition from "buying chat sessions" to "paying for a complete automation production line."