Signal Briefs

The Emergence of Digital Labor Economics

1. The Signal

Early data from agent-driven systems confirms a structural shift:

Agent-enabled users are already generating roughly $1,000–$3,000 in annual token spend.

This is not a projection. This is early-stage reality observed in exmxc’s ADI dataset.

2. The Market Misread

Most market participants are still evaluating AI companies through a SaaS lens:

• users
• subscriptions
• seat-based pricing

This leads to the wrong question:

“How many users does this platform have?”

The right question:

“How much work is this platform performing?”

3. The Correct Lens

AI is not a subscription business.

AI is a metered cognition system.

• tokens = units of work
• agents = workers
• loops = working hours

Revenue is driven by work performed, not users acquired.

The unit of value is not the seat. It is the task.

4. Labor Economics Translation

To make this intuitive, AI should be mapped directly to labor economics.

A traditional personal assistant earning $20 per hour, working 40 hours per week over 52 weeks, has a gross annual wage of $41,600. But gross wage understates the real economic comparison. Once benefits, payroll taxes, and management overhead are included, loaded annual employment cost rises meaningfully.

By contrast, an AI agent operating at today’s early observed spend levels costs a fraction of that amount.

A traditional personal assistant earning $20 per hour, working 40 hours per week over 52 weeks, has a gross annual wage of $41,600. But gross wage understates the real economic comparison. Once benefits, payroll taxes, and management overhead are included, loaded annual employment cost rises meaningfully.

Human cost reflects loaded employment cost including benefits, payroll taxes, and management overhead at roughly 1.3x–1.5x gross wages.

The comparison matters because AI is not merely cheaper software. It is a lower-cost labor layer with different economics:

• 24/7 operation
• parallel execution
• no traditional downtime
• continuously improving capabilities without retraining cost to the buyer

5. The Real Shift

AI is not simply replacing labor.

It is multiplying it.

This is no longer theoretical. At Nvidia’s GTC conference in March 2026, Jensen Huang said he could imagine every engineer needing an annual token budget and floated budgets worth roughly half of base salary, explicitly framing tokens as a productivity input attached to labor rather than a software subscription layered on top of it. That is a meaningful signal that the labor economics framing is entering mainstream institutional thinking.

A single user can deploy multiple agents:

• 5 AI agents at ~$2,000 each = ~$10,000/year
• comparable human staffing cost = well over $200,000/year on a loaded basis

And unlike humans:

• agents can run continuously
• agents can trigger other agents
• agents can create autonomous work loops

6. The True Drivers of Value

This leads to a critical insight:

Users are not the primary economic driver.

The real drivers are:

1. Agent Density (AD)

The number of agents deployed per user or system.

2. Cognition Intensity (CI)

Tokens consumed per task multiplied by task frequency.

3. Loop Persistence (LP)

The degree to which workflows are autonomous and recurring.

7. The Emerging Model

We define:

Cognition Throughput (CT)

CT = AD × CI × LP

This replaces traditional SaaS metrics such as ARR, seats, and basic ARPU.

Note: CT is a directional model pending empirical calibration. AD, CI, and LP are defined indices, not yet directly observable at scale in a standardized public way. The framework is designed to orient valuation thinking, not replace full financial modeling.

8. Valuation Implication

OpenAI’s valuation has moved into a range that many still view through a chatbot lens. Reuters reported in January 2026 that OpenAI was seeking funding that could value the company at about $830 billion, and later reporting said OpenAI projected more than $280 billion in total revenue by 2030 while planning roughly $600 billion in compute spend through that period.

Under the CT model, that valuation is not pricing a simple chatbot company. It is pricing the emergence of a digital labor platform.

If the economic engine is driven by rising agent density, higher cognition intensity, and more persistent autonomous loops, then revenue trajectories can expand far beyond what traditional SaaS models can reasonably construct.

9. The Breakthrough Insight

AI is not sold as software. It is consumed as labor.

And unlike human labor:

• it is elastic
• it is parallel
• it is continuously improving

And unlike human labor, it compounds. Each model generation makes prior use cases cheaper and future use cases more capable.

10. exmxc Perspective

We believe this represents the early formation of:

Digital Labor Economics

Where:

• tokens replace hours
• agents replace workers
• systems replace organizations

11. What Comes Next

This Signal Brief introduces early concepts that will be expanded into a formal framework:

• Tokenized Cognition Model (TCM™)
• Agent ARPU (A-ARPU™)
• Valuation Implied Cognition Load (VICL™)

Final Line

AI doesn’t sell software—it sells labor at near-zero marginal cost. And the companies that scale agents will capture the largest share of global work.

Read the Agent Experience Integrity (AXI) Framework

Read the Agent Readiness Index (ARI)- Tech 100 Report

Read the Lexicon - Digital Labor Economics (DLE)