Blockchain Capital partner Kinjal Shah says the most important effect of AI may not be job replacement, but a deeper rewrite of the basic unit of labor itself. Referencing Sam Altman’s 2024 prediction that a one-person billion-dollar company could emerge in the near future, Shah argues that AI changes production by allowing scale along the dimension humans have always been constrained by: time. Once intelligence is no longer bottlenecked by sleep, working hours, or geography, the way value is created and coordinated begins to change.
In the scenario she outlines, one AI agent could delegate a task to another, receive the output, and settle payment in USDC on-chain within a few hundred milliseconds, all without a traditional intermediary. Shah’s core point is that much of today’s debate is focused on whether AI will take jobs, while missing the more structural question: what happens when labor is no longer organized primarily around jobs, roles, and employment relationships, but around individual tasks that can be priced, assigned, verified, and settled directly?
Lower coordination costs have historically reshaped the firm
To explain that shift, Shah points back to Ronald Coase’s 1937 essay “The Nature of the Firm,” which argued that firms exist because coordinating activity through markets can be more expensive than organizing labor internally. In her reading, every major labor transition in history has followed from a drop in coordination costs. When it becomes easier to find work, manage work, and pay for work, the boundary of the firm moves. Activities that once had to sit inside an organization can move outside it.
She describes the craftsman era as a distributed production model sustained by multi-node supply chains, with each artisan retaining part of the value stack. The Industrial Revolution compressed that structure into factories, where coordination under one roof enabled centralized capture of production value. Later, the internet and mobile computing reduced matching and coordination costs again, enabling the gig economy and creator economy. Platforms such as Uber, DoorDash, YouTube, Instagram, and Substack made it possible for individuals to take on work that previously required studios, publishers, agencies, or tightly managed institutions.
The “bridge class” proves demand before infrastructure captures value
Shah argues that before a new infrastructure layer can absorb most of the economic value in a labor transition, a “bridge class” usually appears first. Craftsmen showed that distributed production could work before factories captured value through centralization. Creators proved individuals could build audiences and monetize at scale before platforms became the Schelling points for attention and revenue. In each cycle, the bridge class carries the early risk, validates demand, and demonstrates that a new labor model is viable. Once infrastructure matures, a new institutional layer consolidates the upside.
In that framing, the gig economy and creator economy were recent bridge classes. They showed that work can be decomposed, distributed, and compensated outside the traditional employer-employee relationship. But they still depended heavily on platforms to package the economic activity. Payments still ran through Stripe, PayPal, or bank accounts; distribution flowed through large content networks; matching remained platform-mediated. So while coordination costs fell, they did not disappear. The underlying payment and identity rails still assumed that both sides of the transaction were human.
Programmable labor plus programmable money changes the unit of work
Shah says the next transition depends on two infrastructure layers arriving at the same time: programmable labor and programmable money. The first is the emergence of AI agents as a new class of labor participant. These agents are not limited by working hours, headcount, or location, and they scale with compute rather than hiring. A top-level agent can break down a task, assign specialized subtasks to other agents, evaluate outputs, and trigger the next step in the workflow, all without human intervention.
Under this model, the fundamental unit of labor is no longer the role, the workday, or even the full deliverable. It becomes the task itself. Historically, humans bundled tasks into jobs, bundled jobs into professions, and bundled professions into firms because that was the most workable organizational format available. But if individual tasks can be directly priced and directly assigned, bundling stops being a structural necessity and becomes a choice.
The second infrastructure layer is stablecoins. Shah notes that stablecoins already represent an asset class of roughly $300 billion, and that multiple institutional forecasts see that figure potentially growing to $2 trillion in the coming years. In her view, stablecoins compress the full payment stack into a single programmable transaction. That matters because the gig economy never fully unbundled labor: on both sides of the transaction, workers and platforms still depended on banks, card networks, and payment processors built for ongoing relationships between known human counterparties.
Stablecoins, by contrast, may be the ideal payment rail for a new labor class made up of agents. One agent can pay another based on output, in amounts as small as fractions of a cent, with settlement in roughly 500 milliseconds and without opening an account, issuing an invoice, or relying on a middleman. That creates the possibility of a production line with no formal organization behind it: no company, no payroll structure, no HR department, just a sequence of tasks being dispatched, executed, priced, and settled at machine speed.
Meta, AWS, and Poncho point to early machine-to-machine commerce
Shah cites several early examples to support the claim that this infrastructure is already moving into production. Meta has recently begun distributing USDC on Polygon and Solana to creators, while AWS has launched AgentCore with support for stablecoin micropayments aimed at commerce between software agents. She presents these moves as signals that large technology companies increasingly view stablecoins as the settlement layer for the next generation of digital economic activity.
She also highlights Poncho, a product from Merit Systems that gives AI agents their own wallet. With it, agents can cross paywalls, access premium tools, and pay for services autonomously, while only paying for the exact usage required. Poncho integrates protocols such as x402 and MPP, embedding payment authorization directly into HTTP requests. In practice, that means an agent can see a price, pay, and immediately unlock access to a dataset, an API call, or a compute resource necessary for a specific task.
For Shah, that represents a different model for how value can move across the internet. Instead of subscribing to a broad package of services that may or may not be needed, an agent can purchase the precise input required to complete one concrete assignment. Earlier waves of the web experimented with micropayments, but those efforts struggled because credit card fees made very small transactions uneconomical, and the internet lacked native payment rails. Stablecoins on networks such as Solana and Ethereum now allow near-instant settlement for fractions of a cent, finally aligning pricing granularity with work granularity.
Firms may survive as orchestration layers rather than labor containers
If more work is completed by agents paying other agents on a per-task basis, Shah argues, firms will not need to internalize every function. Their primary advantage will shift toward defining objectives, specifying quality standards, and combining outputs into a whole that is more valuable than its parts. In that sense, the company of the future may look less like a container that houses labor and more like an intelligent orchestration layer built on top of a global market for programmable labor.
She extends the same logic to the creator economy. Peer-to-peer tipping has had limited success, and platforms such as Clubhouse and Farcaster illustrate its constraints. But micropayments are much more natural in machine-to-machine interactions, where small transactions do not carry social awkwardness or reciprocity expectations. If AI agents become major consumers of digital content, then the subscription model and hard paywall structures that have dominated the internet may gradually give way to metered access executed automatically by software.
Shah concludes that as AI-generated content floods more channels, the premium on human judgment, taste, and craftsmanship is likely to rise rather than fall. The most interesting business models, she says, will emerge at the intersection of human taste and machine execution. In an agent-driven economy, the human role becomes one of rebundling labor: deciding what should be delegated, how outputs should be evaluated, and how separate tasks can be assembled into systems that generate compounding value over time. Firms do not disappear in that future, but their function changes fundamentally.