Microsoft CEO Satya Nadella recently put a name to something many CFOs already recognize, but can't yet measure. The future of every company, he said, will be defined by two things: human capital and token capital. The organizations that figure out how to grow both together will win. The ones that can't productively manage either (or both together) will fall behind fast.
Nadella's framing is sharp. From the beginning, companies have invested in people as a core driver of growth. They have hired employees, trained them, and measured the work they produced, delivered in return for salaries and benefits. But AI has introduced a new kind of value, and it’s accounted for in tokens. Companies are just now sharply aware of this, as more providers shift from flat (or mostly flat) billing per usage tier to itemized billing per token.
Token capital—which is actually a shorthand for token-powered capital—is the value an organization builds when AI use becomes repeatable, measurable, and tied to work. It includes prompt responses, improved workflows, improvements against various measurements and benchmarks, model outputs, and productivity gains. Tokens are the operating input. The return depends on what those tokens produce. The gap between cost and output is where token-powered ROI is created or lost.
"The future of a firm, at a foundational level, will have human capital and token capital. For the token capital, they need their own hill-climbing machine."
Satya Nadella, Microsoft CEO — Stratechery Interview, June 2026
Here’s the problem most companies face right now: they’re consuming tokens at scale, but they don’t even know exactly what that usage is costing them, let alone whether they are creating token capital.
Recent reporting and Larridin scan data point to the same pattern: AI spend can scale faster than finance, IT, or department leaders can explain it.
|
For Uber's engineering team to burn its entire 2026 AI budget |
Unplanned AI spend at one healthcare enterprise before finance understood the cause |
IT leaders hit with AI charges they never budgeted for |
Of enterprise AI spend unattributed on average, per Larridin scan data |
The stories above are not outliers. They’re what happens when organizations scale AI use faster than they scale AI measurement. Token spending compounds quickly. A single active agent running around the clock can burn through a team's quarterly budget in weeks. Nobody planned for it, and nobody caught it. That is the AI spending crisis.
Many companies are trying to close the spending gap with point solutions. There are token-level observability tools, engineering productivity trackers, and FinOps platforms being repurposed for AI. Each category captures one part of the problem. None of these solutions shows spend, adoption, workflow context, and outcomes in one view.
Here’s a quick look at what several current tools actually covers:
|
Tool |
What it does |
Gap |
Strength |
|
Pay-i |
GenAI ROI at the use-case and token level. Strong KPI mapping: CSAT, revenue, time saved. |
Engineering-layer only |
Token-level precision |
|
Worklytics |
Workforce analytics for HR and IT leaders, with AI adoption benchmarks. Ingests 25+ productivity tools. |
No token cost visibility |
Benchmark data |
|
Mavvrik |
FinOps-style AI cost governance. Tracks token costs, GPU spend, SaaS consumption. |
Infrastructure focus only |
CFO-ready cost model |
The issue is that each tool only focuses on one layer of the business: engineering, HR and workforce analytics, or infrastructure cost. These tools fall short when it comes to the enterprise-wide question: what are we spending, where is it happening, and what are we getting back; per team, per department, and across the whole organization?
That is the gap Larridin was built to close.
Larridin is an AI usage and impact measurement platform that connects spend, adoption, workflow context, and outcomes across the organization.
HOW LARRIDIN CLOSES THE GAP
One platform. Every dollar. Every team. Every outcome.
Engineering work may be the easiest to measure because it already leaves a trail: commits, pull requests, reviews, and cycle time. But even in engineering, the parts of the job that AI takes on are only a part of a broader workflow. The challenge is even harder in the rest of the company, where there’s usually no equivalent of a pull request. Larridin addresses that gap by looking at repetitive task patterns in workflow tools, comparing time-on-task before and after AI adoption, and mapping that data against token spend to estimate ROI by role and department.
Larridin is the only platform that comprehensively accounts for spending. And Larridin uses that data to push further, into management of what Nadella has named token capital.
Nadella’s concept is clearly useful at the attention-getting level. But it only matters over the long term if companies can measure it. Doing so requires data infrastructure that most companies don’t have yet. A real token capital picture includes, at minimum:
Larridin's Token Spend & Insights dashboard surfaces all of this in one view. In a typical first scan, it finds an average of 47 orphaned agents, and 18% of agent spend with no owner attached. These findings alone tend to focus the conversation.
This distinction matters. Token spend is what you paid. Token capital is what token spend builds.
Nadella's argument is that companies accumulate token capital as a competitive asset by creating internal benchmarks, fine-tuning model weights, and developing proprietary reinforcement learning environments. The firms that invest in AI systematically and measure what they get for it build a compounding advantage. The ones that just spend tokens without tracking outcomes are burning capital that they will never recover.
This is why measurement is not optional. Without visibility into what token spend produces, you cannot make the case to the board that your AI investments are building anything. You’re just running a cost center with a very large, very unpredictable bill.
Token capital is the value an organization builds when AI use becomes repeatable, measurable, and tied to proprietary work. It includes prompts, workflows, internal benchmarks, model outputs, and productivity gains. Microsoft CEO Satya Nadella recently said companies will need to manage token capital alongside human capital in the AI era.
Token ROI compares the cost of token consumption with the measurable value produced.
Token ROI = value produced ÷ cost of token consumption
Key metrics include cost-per-task, time-on-task before vs. after AI adoption, error or redo rates, output quality, and throughput lift per role. Platforms like Larridin connect token spend to output metrics at the role, team, and department level.
AI spend comes from multiple sources that aren’t connected: SaaS seat licenses, cloud model API calls, coding agents billed per token, and gateway usage. Each measures costs differently. Without a tool that reconciles all of them, finance teams work from an incomplete picture. The problem compounds because autonomous agents can run around the clock, creating variable costs that are difficult to forecast or attribute.
Tokenmaxxing means rapid, often unchecked growth in AI token consumption across an organization. At Google I/O 2026, Sundar Pichai joked that Google’s surging AI usage could be described that way. The business risk isn’t high token use itself. It’s the lack of visibility into cost, ownership, or outcomes.
Pay-i focuses on token-level ROI at the engineering and product layer. Worklytics focuses on HR and workforce analytics, including AI adoption benchmarks. Mavvrik focuses on FinOps-style cost governance for AI infrastructure. Larridin connects AI spend, adoption, workflow data, developer output, and business outcomes across the entire enterprise.
Not sure what token capital you’re creating? Larridin gives you the full picture: what you are spending, which teams and tools are driving it, and what you are getting back.