Coinbase CEO Brian Armstrong shared details of a strategy that allowed the company to nearly halve its artificial intelligence costs while simultaneously experiencing exponential growth in token consumption. This is not the result of strict limits or bureaucratic restrictions, but rather a consequence of smart engineering optimization: default model settings, request routing, and aggressive caching.

Armstrong notes that engineers are free to choose any model, but default settings are crucial. Coinbase is experimenting with setting open-weight models like GLM 5.2 and Kimi 2.7 as defaults through an internal gateway. Notably, 91% of employees never hit limits, so the company chose to reduce costs rather than tighten quotas.

A key element of savings is intelligent routing. Requests are pre-analyzed and directed to the most suitable model based on cache hits and cost. For example, strategic planning requires an advanced model, but it is excessive for routine execution. Armstrong emphasizes that model selection should ideally be automated by AI itself, not by a human.

Caching is highlighted separately. Cache misses are the easiest way to drive up costs. At Coinbase, all requests are configured to reuse already processed information. In the LibreChat service, the cache hit rate increased from 5% to 60% after proper configuration.

Context savings are also important. Armstrong advises starting new sessions when switching tasks, narrowly limiting file context, and disabling unused tools. The goal is not to spend fewer tokens, but to avoid wasting them.

Deutscher's "Barbell" Strategy

Analyst Miles Deutscher describes a similar approach, calling it "token engineering" and a "barbell" strategy to reduce costs by 50% or more. He entrusts the first 10% of work (planning) to the smartest models like Opus or GPT. The main 80% of routine tasks are handled by a cheaper open-source model. The final 10% (result verification) is again assigned to high-level models. Deutscher has been using this scheme for several months and considers it the best way to reduce excessive AI spending.

Cryptalist Analysis: Coinbase's experience is a clear example that the effectiveness of AI implementation is determined not by budget, but by architecture. Dividing tasks by complexity and maximizing caching are the "quick wins" available to any company, not just giants. The market is transitioning from a race for the most expensive models to an era of their rational application.