Your AI rollout is working, but for the people who need it least

Token spend is having a moment. Companies are posting their usage numbers, frontier model costs are climbing, and everyone is suddenly paying attention to where their AI budget is actually going. But most of the conversation stops at how much - not where it lands, and who it's actually reaching. And when you look closely at who it's reaching, the answer is almost always the same: your best people, not your entire organization.

Here's what most companies are actually doing with AI: they're making their best people faster, measuring the gains, and calling it a strategy. It isn't.

The engineers on Claude Code and Cursor, the operators running Manus and Windsurf sessions, the analysts who live in Claude.ai — these people are genuinely 100xing their output. They're not the problem. The problem is that what they're doing is invisible to everyone else in the company, and that gap doesn't close on its own.

At Dust, we've watched this play out across hundreds of deployments. Gabriel Hubert, our CEO, has a blunt read on it: "The companies that fail don't have a technology problem. They have a deployment problem." The numbers back that up: average AI adoption sits at 15% after six months, and 95% of GenAI pilots fail to deliver tangible value. The technology isn't broken. The allocation is.

Tokens are the currency of AI work. Where you spend them is a strategic decision, whether you treat it that way or not.

The first model is what we call the Individual Model: all token spend goes toward personal productivity tools like Claude Code, Cursor, Manus, and Windsurf. These tools work because they give people real agency to move fast without waiting for someone to build them a workflow. The gains for those individuals are real. But individual gains don't compound into company gains. One person's Cursor session doesn't make their teammate better at their job.

The second is the Company Model: your most AI adept employees still get the best tools and freedom to explore. But instead of concentrating everything there, you put roughly 70% of your token spend toward infrastructure that scales what they know to everyone else.

In practice this means running it bottom-up. Your fraud team, ops team, and sales reps build agents that encode their expertise. Leadership sets the platform, the permissions, and the governance, then steps back. The deployments that fail are almost always the reverse: centralized IT teams building what they think the business needs, with no one adopting it because the people who do the actual work weren't involved.

The measure of success matters as much as the rollout itself. Companies that see the strongest adoption don't start by chasing ROI numbers. They start by changing behavior. One CPO put it plainly after deploying AI across 3,000 employees: "Start with adoption, not ROI. The first six months should be about behavior change." The companies that get fixated on proving return too early end up optimizing for the wrong thing and stalling before the compounding kicks in.

At Dust, we build this way ourselves. Our sales team doesn't do manual meeting prep: agents synthesize account history, prior interactions, and market context into briefs automatically. Our engineers don't context-switch to answer the same internal questions repeatedly: agents handle those in Slack before they become interruptions. The humans stay focused on the work that actually requires them. Everything around that work gets handled.

The companies that have moved this way show the results. Persona hit 80% AI agent adoption across their team. Vanta reclaimed 400 hours per week on QBR prep alone, with adoption spreading well beyond the GTM team that started it. As Shashank Khanna, their Founder in Residence of GTM Innovation, put it: "We used to do the work. Now we build the agents that do it." Neither result came from better tools for the people already ahead.

The question every leadership team needs to answer isn't "are we using AI?" It's "who is it actually reaching, and what's the plan for everyone else?"