Detailed Analysis
Anthropic's traffic data for Claude.ai revealed a notable behavioral shift between November 2025 and February 2026, with the concentration of usage across the platform's top tasks declining from 24% to 19% of total activity. This diversification — described by observers as "spiky" and behaviorally distinct from other traffic patterns — signals that users are engaging with Claude across a broader and more varied range of tasks rather than clustering around a small set of dominant use cases. The change is significant not merely as a usage statistic but as a diagnostic indicator of how the platform's role in workflows is evolving.
A primary driver of this diversification is the migration of coding workloads away from Claude.ai and toward Anthropic's first-party API infrastructure, particularly through Claude Code. Because Claude Code's agentic architecture decomposes coding tasks into numerous discrete API calls spanning many categories, what once appeared as a concentrated coding use case on Claude.ai now manifests as diffuse, distributed API traffic. This architectural shift explains why API task concentration has remained comparatively stable even as Claude.ai usage has spread outward. Simultaneously, augmentation interaction types — patterns associated with validation and learning — showed a slight increase on Claude.ai, while automation-oriented interactions declined in API traffic, pointing to a meaningful divergence in how the two surfaces are being used.
The social media thread surrounding these observations reflects broader user tensions with Anthropic's evolving product strategy. Multiple users directed complaints at Boris Cherny, a prominent Anthropic engineer associated with Claude Code, regarding rate limits on subscription plans, the blocking of third-party automation tools from consumer subscriptions, and dissatisfaction with refund and billing processes. Several users explicitly cited competing models — including those from OpenAI, Google, Deepseek, and open-weight alternatives like MiniMax and Qwen — as viable substitutes, with at least one organization announcing a ban on Anthropic subscriptions in favor of those alternatives. The sentiment underscores a recurring tension in AI platform economics: the gap between what power users expect from flat-rate subscriptions and what those plans can sustainably support given token costs.
The traffic behavior data connects to a broader trend in AI deployment: as frontier models become more capable of autonomous, multi-step task execution, usage patterns fragment and become harder to categorize. Agentic workflows inherently generate more heterogeneous API call signatures than simple prompt-response interactions, which is precisely why concentration metrics drop even as overall utilization grows. Anthropic's Economic Index reporting, which found that nearly half of all jobs had at least a quarter of their tasks performed using Claude, suggests the platform is reaching deeper into professional workflows — but in ways that are structurally diffuse rather than monolithic. The "spiky" traffic behavior noted in the original observation is thus a downstream effect of agentic architectures breaking work into smaller, parallelized units rather than large, synchronous sessions.
This dynamic has significant implications for how Anthropic prices and governs access to its models. The friction visible in the user responses — complaints about message limits, blocked automation, and subscription value relative to competitors — reflects the difficulty of designing consumer tiers for a model that is increasingly used in agentic, high-volume contexts rather than conversational ones. The decision to restrict third-party tools from consumer subscriptions while expanding first-party agentic products like Claude Code represents a deliberate attempt to capture the value of automated usage within a controlled commercial framework, but it risks alienating the developer community that has historically been central to Anthropic's growth. Whether that tradeoff proves sustainable will depend substantially on whether Anthropic's first-party tooling can match the flexibility that independent developers previously built for themselves.
Read original article →