Thomas Robinson is the Chief Executive Officer of Domino Data Lab.
The growth of foundation model providers depends on misaligned incentives. These companies profit every time you generate more code, spin up more agents and ask LLMs more questions. They have a huge incentive to encourage token consumption, and they don’t care whether it moves your business forward. Think of it like potato chips or doomscrolling: Both are engineered to be consumed excessively and mindlessly. Before you know it, you’ve spent hours staring at your phone or eaten your way through an entire bag of chips.
Similarly, mechanisms that encourage hyperconsumption are built directly into foundation model products. The “Would you like me to keep going?” prompts, ever-expanding context windows and verbose-by-default answers are all by design. The industry has taken the token consumption dial out of the user’s hands and built a self-reinforcing loop to keep usage high.
And for a while, “tokenmaxxing” was the goal for users, too. Companies encouraged consumption in the hopes of maximizing productivity and innovation. Then the bills came. In response, many organizations have capped usage to contain out-of-control spend. But that’s just treating the symptom, not the cause. To see value from AI, we need to go back to basics. We need to start “valuemaxxing” by tying every dollar of AI investment directly back to the handful of core business metrics that already define success.
Here’s what that looks like in practice.
Make business value (not token spend) your North Star.
There’s no single, clear proxy for AI value. But one thing is for sure: The number of employees using AI, licenses deployed or tokens burned won’t tell you anything meaningful about how AI is impacting your business. The only defensible approach is to go back to the handful of metrics that already define success in your organization—things like revenue, cycle time, error rate or whatever your board watches—and closely track the impact AI has on them. Every AI initiative should map back to a tangible business metric.
Build fewer, higher-value systems.
Every business function has just a few metrics that truly matter. The goal should be to build a select number of high-value systems to address those outcomes. What many companies end up doing is inadvertently churning out an endless stream of pilots and AI slop.
Take a “work-backward” approach: Start with a genuine business problem, assess whether AI is really the right tool to address it and then—and only then—should you build. Not following this approach is how you end up with things like the useless website chatbots consumers have come to despise. Save the “let a thousand flowers bloom” philosophy for genuine R&D, not your core operations.
Implement guardrails instead of spending caps.
Token ceilings are necessary to save you from a runaway bill, but they can’t guarantee your employees use AI mindfully. They’re a reactive, short-sighted solution to a deeper problem. Instead, implement shared AI platforms, an approved model catalog and reusable services so every team doesn’t feel the need to go off and build their own tools.
AI spend should be treated like an expense account. Employees don’t get free rein in what they expense to their corporate credit card, nor should they have the ability to “vibe code” a custom app for a problem your existing tools already solve. Historically, companies have always rationed emerging tools and technology to the people who needed them most (e.g., field sales teams got laptops first), and access to AI should follow a similar logic—enforced by architectural guardrails.
Make your people the heroes of AI again.
AI is a labor multiplier, not a labor replacement. Treating it as the latter is how companies like Ford and Klarna ended up scrambling to rehire a portion of their workforce.
So, invest in your people accordingly. As a baseline, every employee should have access to an AI productivity tool. From there, a significant chunk of your AI budget should go toward teams responsible for critical business processes to encourage real value-building, not just “random acts of AI.” Your core development team also needs adequate budget to support AI in the software development life cycle. Finally, R&D teams require some extra leeway to explore freely so they can accelerate discovery. Whatever your approach, have a clear playbook when it comes to empowering your workforce with AI so it doesn’t become a free-for-all.
“Valuemaxxing” isn’t just a buzzword—it’s the only strategy that was ever going to work. It’s time to stop chasing tokens and endless AI pilots and start walking backward from real business metrics. AI can’t create business value on its own, so invest in your people and optimize for better outcomes, not just output.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?