The AI revolution has entered a new phase, one where the all-you-can-eat buffet is over, and the focus shifts to counting calories. This shift is not just about cost-cutting measures but also about a paradigm shift in how companies approach AI. The era of unbridled AI usage, fueled by cheap tokens and flat-rate billing, is giving way to a more measured and deliberate approach. This transformation is particularly evident in the coding world, where companies are now rethinking their AI strategies and imposing limits to manage costs and optimize usage.
One of the key players in this shift is Coinbase, a crypto exchange that has seen its AI usage skyrocket since the launch of Anthropic's Claude coding model. Coinbase's infrastructure head, Rob Witoff, noted that the company's internal usage of AI started to 'go parabolic' after the launch of Claude's improved coding model, Opus 4.6. To manage this surge in usage and rising prices, Coinbase has implemented a sophisticated system of weekly price caps based on job level and role. This system not only helps control costs but also encourages employees to be more mindful of their AI usage.
The shift towards more controlled AI usage is not limited to Coinbase. Other companies, including Walmart, Amazon, and Accenture, have also imposed usage limits and backed initiatives to standardize AI budgeting metrics. This trend is not just about cost-cutting; it's about creating a more sustainable and responsible approach to AI adoption. The novelty of AI has worn off, and hard-nosed utility has stepped in, according to Niranjan Krishnan, the head of AI solutions at FPT Americas.
The high stakes of AI adoption, particularly in the context of upcoming IPOs, are prompting scrutiny and a reevaluation of spending. Salesforce CTO Parker Harris, for instance, acknowledged that the company has fully opened the floodgates for spending on Anthropic tools but is now looking to find a balance that doesn't divert too much money to the rising startup. This balance is crucial for public companies, which must consider the impact of AI spending on their investors and the broader market.
The economic proposition for AI providers has also changed. Tokens have become cheaper due to Nvidia's chip innovations, but the popularity of AI tools and new agent-based setups have made it unsustainable for providers to subsidize heavy users. GitHub, for instance, has shifted to a usage-based pricing model, which has been met with complaints from developers who are now facing significant increases in their AI code expenses. OpenAI and Anthropic have responded by promoting their newest model releases as more 'token-efficient' and offering options to complete non-urgent tasks at lower prices.
The coding craze is still in its infancy, and the 'tokenmaxxing' uproar, sticker-shock pricing, and see-sawing behavior by companies are all symptoms of a sudden paradigm shift. Companies are proceeding with caution, trying to avoid holding back on the gas while also trying to understand the tangible impacts of AI. Salesforce, for instance, is rolling out measurements for both customers and internal engineers to better understand the impact of AI. This includes an Effective Output score, which will help avoid further surprises and enable companies to forecast and manage their AI spending more effectively.
In the meantime, some companies are offloading basic work to less advanced AI models, either from American companies or from Chinese firms like Deepseek and MiniMax. This shift is driven by the need for cheaper models that can handle basic tasks without the high costs associated with cutting-edge models. The calculus for companies is simple: what's the return on investment for AI, and what's the right spend for that return?
The AI revolution is far from over, but the focus on counting calories is here to stay. As companies navigate this new phase, they are learning to balance the benefits of AI with the need for responsible and sustainable adoption. The future of AI is not about unbridled usage but about finding the right balance between innovation and cost-effectiveness.
