# AI@Work: Tokenomics is the new headcount—plus 4 more trends to watch > Source: > Published: 2026-06-04 18:30:19+00:00 Two years ago, every conversation about AI started with the same question: can this technology actually do valuable work? Today, the viability of AI has been proven. The question that’s now top of mind is, how do we lead through the transformation? The world of AI is moving too fast for any organization to figure out alone. That’s why we recently convened 250 of our customers operating at the leading edge of AI transformation in a forum called the Copilot Summit. Our goal was to shorten the distance between the teams building Copilot and the leaders using it. What struck me most was how universal the hardest challenges turned out to be, and how consistently they pointed to the same reality: AI returns are determined by the decisions leaders make, not by the technology they buy. Five takeaways from the leaders in that room stood out. Trust in AI is specific, not general Knowledge work inefficiency demands a redesign The system matters more than the model Tokenomics is the new headcount Enterprise software must now earn the right to exist **Trust in AI is specific, not general** Trust is built on specifics. Not broad confidence in a technology, but confidence in a particular system doing a particular job. You wouldn’t expect a child to fly a plane. That isn’t a statement about aviation, it’s about fit between capability and task. Trust in AI works the same way. Trevor Noah made that argument in a session that reframed how a lot of people in the room were thinking about the problem. He pointed to Johns Hopkins cancer research as the clearest example of what appropriately scoped AI looks like: a system trained on a single dataset, focused entirely on minimizing unnecessary biopsies for breast cancer patients—not writing poems, not giving directions, not doing anything outside its defined purpose. The specificity is exactly what made it trustworthy. The contrast, he noted, is an AI agent like the one some users discovered inside Hertz’s customer service flow—where if you dug far enough, you could prompt it to write code. The system had no edges, and without edges, there’s no basis for trust. Three conditions build that trust over time: consistent performance, a working understanding of how the system functions, and accountability when something goes wrong. It’s easy to underinvest in that last one. Trevor’s example was commercial aviation—the reason people get back on planes after a crash isn’t optimism, it’s the FAA report, the public accounting of what went wrong, and the demonstrated consequences for the companies at fault. As agents move from generating content to taking action, the accountability infrastructure needs to exist before the failure, not after it. **Knowledge work inefficiency demands a redesign** Knowledge work has been largely ad hoc for decades. Even domains with apparent structure (legal, finance, accounting) run on vague workflows, inconsistent outputs, and goals that live in people’s heads rather than in any system. [Charles Lamanna](https://www.linkedin.com/in/charleslamanna/), who leads our Copilot and Agent Platforms, made the case that the same structural shift that redesigned manufacturing is coming to every knowledge-work function: measurable steps, deliberate trade-offs between human and machine labor, tracking of outcomes rather than activities. [Katy George](https://www.linkedin.com/in/katygeorge1/), who leads workforce transformation at Microsoft and oversees our own journey as Customer Zero, brought that argument to ground level. Our first attempt to roll Copilot into our own sales force fell short. It wasn’t because the technology didn’t work, but because we treated it like any other product launch. Adoption metrics moved, but outcomes didn’t. The result comes from redesigning the work the tool sits inside, not from giving people access to it. **The system matters more than the model** [Ryan Roslansky](https://www.linkedin.com/in/ryanroslansky/) spent years watching LinkedIn members navigate some of the most consequential decisions of their careers—where to work, what to build, whether to stay. He has argued that the professionals pulling ahead aren’t necessarily the most credentialed; they’re the ones who understand which systems they’re operating inside. He brought that same instinct to his session at the Summit, and it reframed how a lot of people in the room were thinking about AI strategy. For the first wave of AI deployment, the model was the decision. Organizations asked which one to use and treated the answer as the work. Then it became clear that the model alone wasn’t enough—the harness around it mattered just as much: the data it could access, the context it was given, the infrastructure it ran on. And the elements required to deliver real value keep multiplying. What matters now is not any one element, but how deliberately the system is constructed end to end. Selecting a model is the starting point. Building the system around it is the work. What that means in practice is that AI capability is increasingly a construction project, not a procurement one. The organizations pulling ahead aren’t finding better models—they’re building more deliberately around them, assembling the right elements painstakingly enough that the system can actually deliver. Technology diffuses through the work, not around it. The ones getting there are embedding AI deeply enough that the technology recedes, and the work comes forward. **Tokenomics is the new headcount** Token economics—or tokenomics, as it’s becoming known—is worth defining clearly, because it reframes something fundamental about how leaders make decisions. When AI tools first entered organizations, leaders evaluated it against their IT budget—a number they understood, in a category they already managed. Tokenomics works differently. The relevant comparison is the cost of a human doing the same work, not a software line item. Now that AI is capable enough to do real work, every leader has to answer a question they’ve never had to answer before: should a human do this, or should an agent? That calculation runs across quality, time, and cost. And the cost piece is moving fast, as AI models and systems become better, faster, and more efficient. What it costs today won’t be what it costs next year, or even next quarter. The allocation question that follows is immediate and concrete: who gets tokens, how many, and for what work. Think of it like managing headcount: the same deliberateness, the same trade-offs, the same accountability for outcomes. Most organizations don’t yet have the infrastructure to make those decisions well. The ones that build it now, and recalibrate as the economics shift, will have a meaningful advantage over those treating token costs as a static number. The tokenomics of AI work What a task costs to run today won’t be what it costs next quarter. Budget using current token prices, and your decisions about deploying AI could leave value on the table. **Enterprise software must now earn the right to exist** “The era of ‘I use this kind of crappy thing because I’m forced to use it’ is kind of over.” [Jacob Andreou](https://www.linkedin.com/in/jacob-andreou/) said that in his session, and it landed. Andreou joined us from Snap, where he led product and growth, and now runs Copilot end-to-end. He was naming a shift that has been building for more than a decade. Starting in the early 2010s, consumer products began crossing into work. The iPhone was the first major inflection point: people experienced something exceptional in their personal lives and carried that standard into the office. Enterprise software didn’t have to compete on that basis for a long time. Decisions were made by IT and finance, deployment was mandated, and the gap between what a product promised and what employees experienced was treated as a cost of doing business. AI closes that gap or it fails. People now have strong reference points for what a great AI experience feels like. They use these tools in their personal lives, and they bring that standard to work. The organizations that hold their AI investments to consumer-grade scrutiny will build differently, buy differently, and measure success differently. **What it all means for leaders** Across all five takeaways, the same thing kept surfacing. The technology is a constant—what varies is the quality of the decisions made around it. How trust gets designed in. Whether the work gets genuinely redesigned. What sits beneath the model. How a new kind of resource gets allocated. What standard an investment has to clear. None of those are technical questions. They’re organizational ones, and they require someone to answer them deliberately. That’s what the organizations pulling ahead have in common. For more insights on AI and the future of work, [subscribe](https://www.linkedin.com/newsletters/ai-at-work-7114636182952562688/) to this newsletter.