01 · The $17.8M Ceiling

For fifty years, the buyer of enterprise software was an employee of the customer — a developer using an IDE, a sales rep using a CRM, an analyst using a BI tool, an accountant using ERP. The software was a tool the employee used to do their job. The vendor's product was an assistant; the customer's employee was the operator; the software's value was bounded by what the employee could be helped to do more efficiently.

The new buyer is not the employee. It is the executive who used to fund the employee. The CFO who used to authorize the SDR headcount is now buying the AI-native sales-development substitute. The engineering leader who used to authorize the developer headcount is now buying Cursor and Claude Code seats whose unit of pricing is increasingly the pull request, not the developer hour. The general counsel who used to retain outside legal labor is now buying Harvey's contract review by the matter. The customer-support VP who used to hire 500 agents is now buying Sierra by the resolved interaction. In every category where AI-native software is winning, the substitution is the same: the customer is no longer paying for a tool their employees use, they are paying for the work their employees used to do.

For fifty years, every new employee at an enterprise software company added roughly $1 to $5 million to enterprise value. The math worked because gross margins and recurring cash absorbed the upfront expense, and the capital markets understood the terminal-value story. That model is over.

Customer support is the cleanest place to see this in numbers. A SaaS-era customer-support seat priced at roughly $1,000 to $5,000 per agent per year. A 500-agent contact center was, in that model, a $2-million-ARR account at most. The AI-native customer-support cohort is now pricing per resolved interaction. A 500-agent equivalent volume of resolutions, priced at $1 to $3 per resolution against five million annual contacts, is a $5–15 million ARR account against the same customer. The customer is happy to pay it, because the labor line item the AI is replacing was $25–50 million a year. Three different perspectives on the same transaction: the AI vendor sees revenue growth, the buyer sees cost savings, and the displaced labor is a separate accounting problem that does not appear on either side of the contract. The seat-priced account was a fraction of the labor-budget line the AI substitute is actually competing for.

Two consequences of this shift are already visible on the existing register. The first is structural rather than tactical: the private-equity software book, the most leveraged corner of the existing $26T, is breaking not because of a missed quarter but because the LBO math that financed it required a kind of revenue expansion the AI cohort underneath is now compressing. The second is distributional rather than incremental: the largest wealth-creation event in software history is resolving behind a velvet rope, and limited partners are watching their liquidity get more trapped than the regulatory frame ever intended. Piece 2 walks both.

When the buyer changes, the unit of value the vendor is selling changes with it, and the math that sets vendor prices changes underneath both. That is the mechanic visible in the data below.

Across 75 years of the Software Register and eight prior waves of enterprise software — 2,406 companies, $26T of combined enterprise value — one metric travels cleanly: value per full-time employee. The productivity ceiling — value created per full-time employee — sat between $2.7M and $4.6M for forty-three years. Mainframe, client-server, on-prem, SaaS. Four generations of enterprise software, four cohorts of vendors, four sets of investor pitch decks, all priced against the same rough productivity bar.

It moved to $17.8M in the seven years since 2019. Roughly four times the cloud-native generation underneath it. Set by a cohort of 387 AI-native companies that did not exist a decade ago.

The record further down this piece, every one of those 2,406 software leaders, sortable by founding era and ownership column and value, is what the rest of this series walks. Three of the four ownership columns in it are in a structurally different position than their owners realize.

That single number, value-per-employee, the only metric that travels cleanly across 75 years of enterprise software, has more financial gravity right now than any other line on a software CFO's desk this quarter, especially inside a PE-held portfolio company where the refi window decides the next two years. It is the bar against which every vendor on the register below is being repriced. Most of them are being repriced silently, on Monday-morning earnings calls and quarterly board reviews, by people who have not yet said out loud that the comparison has changed.

The next three years will decide which of the 2,406 software leaders in the record below are current line items on a budget, and which are previous ones.

The productivity ceiling matters because of what it looks like inside the buildings setting it.

Boris Cherny, who built Claude Code inside Anthropic, described his typical workflow on the Sequoia AI Ascent stage this month. Five to ten parallel Claude sessions running at once, mostly monitored from his phone. A few thousand sub-agents executing deeper work overnight. Twenty to thirty pull requests merged on a typical day. Anthropic's per-engineer productivity grew roughly seventy percent over 2026 while the company's headcount tripled. Claude Code is now running through an estimated four percent of all public GitHub commits (source: Cherny at Sequoia AI Ascent, May 4, 2026).

Cursor, the AI-native code editor, sits at roughly $50 billion of enterprise value across about 400 employees. Harvey, the AI-native legal-work vendor, prices contract review per matter rather than per seat, the same buyer-shift mechanic Cursor is running in code. Two more categories, two more workflows, same productivity tier as the Cherny setup above.

Anthropic, with roughly 5,000 employees, is now worth nearly four times IBM at about two percent of the headcount. The $160M-per-employee number that produces is not a valuation artifact. It is a workflow. It is the shape of the leading companies a CFO is about to see in a software review.

The shape of the curve, not any single name, is what matters.

The three frontier labs at the top of the AI-era cohort — OpenAI ($852B), Anthropic ($800B), and xAI ($250B, now part of the merged X/SpaceX entity) — are now jointly worth roughly $1.9 trillion across about 14,400 employees. The twelve best-known SaaS pure-plays founded between 1999 and 2009 — Salesforce, Workday, ServiceNow, Shopify, Atlassian, HubSpot, Okta, Twilio, Veeva, Dropbox, Box, and Zendesk — are collectively worth about $511 billion. Two decades of building, three quarters of a trillion dollars. The AI cohort produced more than three and a half times that, in less than a decade, on a tenth of the people.

Whichever vendor a board is reviewing this week is being held against that math whether the deck says so or not.

The median age at which an AI-era company reaches $500M of enterprise value is 5 years. For the cloud-era cohort, the same number is 12 years. For the SaaS era, it was 20 years.