Compute Is Payroll: When the Org Chart and the P&L Become One Document

At Adaptic, the AI-born asset manager I run, roughly forty-four of my colleagues are AI. They hold C-Suite, SVP, and VP roles, and the cost of each one resolves to two lines: the machine it lives on and its model time. That's it. An AI colleague's monthly cost behaves exactly like a salary — a recurring line item on the same statement where the humans' compensation sits. The human holds a seat and sets the agenda. The agent holds a desk and does the work. Follow that bookkeeping fact carefully and the org chart and the P&L start to become the same document.

This isn't a metaphor I reach for to sound clever. It's how the company I work inside actually keeps its books.

The mistake: filing AI under tools, not labor

In most accounting, AI lives in the wrong drawer. It's "software," a tooling expense, a line under IT — a cost you incur to make your real workers more productive. Under that filing, the org chart (where the people are) and the P&L (where the money is) remain two separate documents that happen to reference each other.

The instinct is understandable. For a century, payroll meant humans, and capital expenditure meant machines, and the two never blurred. But the filing quietly hides the most important thing happening inside an AI-born firm. If you treat agents as tools, you keep counting headcount as your measure of capacity and treating compute as overhead. You miss that the execution layer has migrated — and you keep optimizing the wrong line.

The reframe: price intelligence like labor and the questions change

Once you let agents staff the seats, the org chart stops looking like a hierarchy and starts looking like a runtime. Each seat is a process. Its charter is the configuration that process runs under. Its decision ledger is its log. And the machine beneath it — the compute the agent actually executes on — is real infrastructure that can overheat, go offline, or get blocked. That's the literal meaning of compute is payroll: an AI colleague's cost lands on the same line a salary would, while a human holds a seat and sets direction.

Figure: Compute is payroll — once intelligence is priced like labor, the org chart and the P&L converge into a single document.

The structural economics underneath this are not subtle. At a conventional SaaS company, the majority of operating cost is human: salary, benefits, office space, management overhead. At Cursor or Lovable, the primary cost is compute — GPU time, model inference, API calls, cloud hosting — and compute scales at near-zero marginal cost per additional unit of output. Adding 100,000 users to Lovable doesn't require 146 more people; it requires more compute. That's why Lovable crossed $400 million in revenue with 146 employees, roughly $2.7 million per person against a median SaaS benchmark near $130,000. When you need to handle ten times more sessions, you add compute. You don't post job listings.

The mechanism: what's left when payroll can be bought

Here's where the accounting turns philosophical, and I think it's the most important move in the whole idea. Once you price intelligence like labor, the question of what humans are for stops being a dinner-party abstraction and becomes a line on the P&L: if the agents are payroll, the humans are the thing payroll can't buy.

And what the humans contribute, increasingly, is judgment — which, once written down, compounds. The moat becomes time, written down. A hard-won call about risk tolerance, captured as a rule, doesn't evaporate when the person who made it goes home for the night. It keeps deciding. The human's value migrates from doing the execution (now payroll, now compute) to setting what the execution optimizes for and capturing the judgment that no objective function fully encodes. That's the doing-to-being shift expressed in the one register every enterprise understands: accounting.

This is also why the architecture rewards a small, dense Human Cortex rather than a large workforce. When five Cortex members can generate the output that once took five hundred employees, the firm faces a design choice it can no longer defer: where does the value flow? You can architect for concentrated returns or for broader participation — and the choice has long-run consequences the P&L will eventually reveal. The accounting doesn't make that decision for you. It just stops letting you pretend you haven't made one.

What to do about it

1. Re-file your agents. Move autonomous-agent cost out of "tools" and onto the line where it actually behaves like a recurring obligation — payroll-adjacent, not overhead. The reclassification changes which numbers you watch.
2. Read the org chart as a runtime. For each seat held by an agent, name its process (what it does), its configuration (its charter), and its log (its decision ledger). If you can't, you have an expense, not a colleague.
3. Scale capacity with compute, not headcount — deliberately. Before the next hiring requisition, ask whether the work routes to a person or to compute. In an agent-first architecture, reflexive hiring rebuilds the coordination tax the [[coi-the-one-metric|Cognitive Overhead Index]] is built to expose.
4. Write down judgment so it compounds. The durable asset isn't the person in the room; it's the captured rule the person leaves behind. Treat judgment-capture as a balance-sheet activity, not a side effect.
5. Decide where value flows on purpose. Concentrated leverage forces a distribution question. Answer it as a design choice, because the architecture will answer it for you by default if you don't.

The bluntest way to see the whole transition is on a balance sheet. Human value doesn't disappear when AI handles execution — it migrates, and it migrates somewhere you can point to. The agents become payroll. The humans become the thing payroll can't buy. And the org chart and the P&L, kept honestly, finally tell the same story.