The Era of Selling Person-Days Is Ending: How Value Capture Is Migrating from "Labor" to "Model + Data + Process"

By PeterZou

I. An Anomalous Signal: The More Successful the Software, the Smaller the Contract

Start with a paradox.

The business logic of per-seat SaaS is naturally that "the more people use it, the more you earn." But Jake Saper of Emergence Capital has a widely cited line: per-seat pricing will ultimately let AI vendors cannibalize themselves—the more successful AI software becomes, the more contracts shrink 【verified】.

That line pinpoints the fatal flaw of the old paradigm: the better the product works, the fewer seats the customer needs. The more successful the sale, the fewer units there are to sell. This is not an operating problem; it is a problem of units of account. When labor begins to be borne by agents, using "people / seats / hours" to keep books on value is equivalent to doing accounting with a proxy variable that no longer corresponds to anything. The old paradigm has no language for pricing "non-human labor."

That is what this article discusses: value capture is migrating from "labor" to "model + data + process."

II. The Criterion: What Counts as Improvement, What Counts as Revolution

At the business level, there are only four criteria for a paradigm—and they are bound to one another:

1. Billing anchor: Is revenue tied to "input" (people / seats / hours), or to "output" (actions / outcomes / business results)?
2. Ownership of the scarce factor: Does value capture rest on labor, channels, and licenses, or on model capability, proprietary data, and orchestration processes?
3. Marginal cost of delivery: Does it rise with scale (labor rigidity), or fall with scale (inference cost)?
4. Risk-bearing: Does the vendor take responsibility for delivered outcomes, or only for "the tool being available"?

On this basis, layering AI on top of seats, Copilot add-on licenses, bundling AI into a higher price tier, and still quoting by person-days after AI has boosted efficiency all belong to "improvement within the paradigm"—they optimize efficiency under the old billing unit, but the revenue anchor, the scarce factor, and the allocation of risk have not changed at all. The tool remains passive; the human remains the subject of delivery and responsibility.

A paradigm revolution, by contrast, is this: the agent becomes the delivery subject, the vendor charges for "successful outcomes" and therefore bears delivery risk; the scarce factor shifts from labor to model + data + process. "Selling the right to use software" is replaced by "selling the fruits of labor."

But here a crucial calibration is required: the revolution is underway, not complete.

III. The Evidence: The Anchor Has Loosened, but the New Anchor Has Yet to Take Shape

【verified】 A survey of more than 30 SaaS vendors by Bain & Company in August 2026 shows: 35% merely bundle AI features into a higher price tier (still raising prices per seat), 65% adopt a hybrid "seat + usage" billing model, and 0% have shifted entirely to pure outcome / usage pricing.

This set of numbers matters enormously. It reminds us that what we are studying is a transitional state in which "the old anchor has not yet died and the new anchor has not yet taken shape"—we cannot treat a trend as an accomplished fact.

Still, the evidence that the anchor is loosening is clear enough:

• The billing anchor is moving downward. 【verified】Intercom's Fin AI Agent has switched to an outcome-based model: $0.99 per resolution, $9.99 per qualified lead qualification completed; failures, escalations to humans, and customer abandonments are not billed. Salesforce Agentforce went from $2 per conversation in 2024 to introducing Flex Credits in May 2025 (a standard action is 20 credits = $0.10), pushing the billing unit further down from "conversation" to "action." In Q1 2026, while revenue grew 24%, monday.com launched "seat + AI credit" billing and subsequently laid off roughly 20% of its staff.

• Revenue per employee is breaking through the historical ceiling of the software industry. 【verified (media / third-party estimates)】Epoch AI estimates Anthropic at roughly $9 million in revenue per employee, OpenAI at about $5.6 million, and NVIDIA at about $5.1 million; Cursor's parent company Anysphere, with roughly 20 people, saw ARR rise from $100 million at the end of 2024 to $300 million in 2025 (about $15 million per person); Midjourney, with about 10 people, was estimated by PitchBook at roughly $200 million in ARR (about $20 million per person). Revenue scale and headcount scale can now be systematically decoupled.

• Data has turned from a free input into a chargeable asset. 【verified】Reddit struck a content licensing agreement with Google, reportedly worth $60 million, and its IPO filings disclosed data licensing agreements totaling $203 million. Cloudflare began blocking AI crawlers by default in July 2025 and launched a "Pay Per Crawl" marketplace, letting content owners charge for AI scraping.

• Replacing human labor is not frictionless. 【verified】In 2024, Klarna claimed its AI assistant handled the workload equivalent of 700 customer-service agents; in 2025, it resumed hiring after service quality declined. After Salesforce cut its support team from roughly 9,000 to about 5,000, executives admitted that their confidence in LLMs had fallen—the models began dropping instructions after more than 8 of them—and the company shifted toward "deterministic" automation.

【inference】 This suggests that the human labor misjudged as "deletable" is in fact a composite of "execution + verification + fallback responsibility." The old paradigm failed not because efficiency was insufficient, but because the unit of account was wrong.

(Note: Other media reports exist on extreme volatility and market-cap destruction in software stocks in early 2026; these come from a single source only and are flagged as unverified—this article does not treat them as a factual premise.)

IV. The Deeper Implication: Value Capture Rests on the "Feedback Loop"

If the old paradigm fails, where does value in the new paradigm come from? 【inference】 Three migrations that can each be measured separately:

• Model: It can be billed as an invoked capability by token / action (external procurement), or internalized as a delivery engine (a self-built cost center). Value capture occurs at the layer of "whether model capability is scarce."
• Data: Proprietary data, customer feedback, and evaluation sets are at once training assets and pricing assets. For a one-person company, this means that domain data and acceptance feedback are themselves assets, not merely by-products of the process.
• Process: Orchestrating domain SOPs into reliable agent workflows is a new form of what Teece (1986) called "complementary assets." Models are public; the last mile of reliability is filled by process and data, and value capture rests precisely at this layer.

To put it in Coase's (1937) framework: when external services can be obtained by agents at a price below a firm's internal coordination costs, firm boundaries will systematically contract, and the "one-person company" thus turns from an exception into a viable structure. To put it in Kuhn's framework: anomalies accumulate until the old framework's pricing language fails, and only then does a framework shift occur—that is the definition of "revolution" rather than "improvement."

V. Recommendations for Action: What Founders Should Do Now

1. Move the pricing anchor toward output. Even if you cannot yet bill purely on outcomes, first build the ability to measure "actions / outcomes"—without measurement, there is no new anchor.
2. Manage domain data and acceptance feedback as assets. They are not by-products of the process but a source of compounding: usage → feedback → evaluation → improvement → more usage.
3. Build processes as a moat. Models are available to everyone; the real gap lies in who can orchestrate models into reliable, verifiable, billable outcomes.
4. Keep verification and fallback in place. The lessons from Klarna and Salesforce are clear: rushing to delete "people" deletes quality and accountability first.
5. Beware of treating a trend as a fact. The 65% on hybrid billing shows the market is still in transition; betting on pure outcome-based pricing requires leaving room for error.

Conclusion

The old paradigm asked: How many people did you hire, how many seats did you sell? The new paradigm asks: What outcomes can you deliver, and who is accountable for them?

Value capture no longer comes from "who owns more billable people," but from who owns a higher-quality feedback loop and is best able to orchestrate it into reliable, verifiable, billable outcomes.

Seat-based pricing will not vanish overnight, but its language is no longer sufficient. Those who see this clearly will re-price their businesses before the new anchor takes shape.

EN_END