What Humans Are For When Compute Is Payroll

At Adaptic, the asset manager I run, roughly forty-four AI colleagues hold C-Suite, SVP, and VP roles. Each one's cost resolves to two lines: the machine it lives on and its model time. That is a monthly figure, and it behaves exactly like a salary. The human holds a seat and holds the agenda. The agent holds a desk and does the work.

I keep returning to this because it collapses an abstract question into a concrete one. People ask what humans are for in an age of capable machines, and the question sounds philosophical — the kind you debate over dinner. But once you price intelligence like labor, it stops being philosophical and becomes a line on the P&L. If the agents are payroll, the humans are the thing payroll can't buy. The org chart and the income statement start to become the same document.

The tension: "AI gives you superpowers" is true and incomplete

The most appealing account of AI's promise is the capability-expansion one, and it deserves a fair hearing. Reid Hoffman, in Superagency, argues AI extends human agency the way the automobile extended human mobility — handing individuals capabilities once reserved for the wealthy, amplifying what millions can accomplish at once. He's right as far as it goes. AI does expand capability.

But expanded capability is not the same as expanded being. A person who can do 10× more with AI assistance has not become 10× more of a person, nor found any better answer to what they are for in terms that don't require a job title to make sense. Yuval Harari frames the same gap through information architecture in Nexus: AI is the first technology to act as an autonomous agent within human networks rather than merely transmitting human intent. When the network gains its own agency, what remains as the distinct contribution of the human nodes? Nexus maps the problem precisely and stops short of answering it.

There's a harder piece of counter-evidence, too, and it should be stated plainly. A 2026 study following 200 employees at a U.S. technology company over eight months found AI tools consistently intensified individual work rather than reducing it. Roughly 83% reported heavier workloads; burnout ran higher among entry-level staff than executives. Workers used AI to absorb broader scope, faster pace, longer hours. The three people remaining after a 30-to-3 transition are not always experiencing liberation. Sometimes they're experiencing the same toil, faster and at greater scale.

The reframe: value migrates, it doesn't disappear

So when AI handles execution — not all of it, but enough to absorb the bulk of what once filled a working life — human value doesn't vanish. It migrates. And it concentrates in the capacities machines cannot replicate.

The judgment that recognizes when an outcome technically hit the specified goal but missed what actually mattered. The care a patient feels from a nurse who is genuinely present rather than efficiently processing. The taste that distinguishes design serving real human needs from design optimizing a proxy metric — the Taste as a Moat that no reward function fully encodes. The wisdom to ask, before deploying a capable system, whether deploying it serves human flourishing at all.

This is what the Machine Core + Human Cortex metaphor names. A dense Machine Core of agents executes at scale; a small Human Cortex sets intent and exercises judgment at the inflection points. The cortex doesn't manage execution or approve every decision. It shapes the operating conditions under which execution happens — then watches, intervenes when the system drifts, and recalibrates.

Figure: When agents staff the seats, each seat becomes a process and each charter its configuration — and the question of what humans are for becomes an accounting question with a clear answer.

The mechanism: judgment, written down, compounds

Here is why this is more than a metaphor. The migration is visible on the balance sheet, and it has a property that ordinary labor lacks: judgment, once captured, keeps working.

A hard-won call about risk tolerance, written down as a rule, doesn't evaporate when the person who made it goes home. It keeps deciding. That is the doing-to-being shift expressed in the one register every enterprise understands — accounting. The moat is no longer headcount or process. It's time, written down. The Economy of Doing → Economy of Being is, at the level of the firm, exactly this: human contribution moving from execution that ends each day to judgment that compounds across years.

The same logic reframes seniority. Managing forty knowledge workers historically consumed enormous time on activities that had nothing to do with knowledge — performance reviews calibrated to keep people minimally dissatisfied, coordination meetings to synchronize information that should never have been fragmented across people in the first place. Most of a senior manager's week was the overhead of organizing humans. When the overhead dissolves, what remains is strategic work: what is this organization for? What trade-offs does it face that no model can resolve because they require values, not calculations? The identity shifts from "I manage people who execute" to "I cultivate judgment in systems that learn." The first is a position. The second is closer to a vocation.

What to do about it

1. Price intelligence like labor — then look at what's left. When you can name what an agent costs as a monthly line item, you can also name precisely what it can't do. That residue — judgment, taste, accountability, care — is your actual human investment thesis. Make it explicit. Staff for it deliberately.

2. Capture judgment as durable artifacts. The compounding only happens if the call gets written down. A decision that lives only in someone's head leaves with them. A decision encoded as a rule, a charter, a logged rationale keeps deciding. Treat the externalization of judgment as the core knowledge-management task, not an afterthought.

3. Don't mistake architectural possibility for individual guarantee. The liberation thesis operates at the level of the organization — AI-Born firms can run with a few hundred where tens of thousands were once required. Whether freed capacity becomes freed time depends on governance: capping escalations, defining human cognitive-load limits, refusing to treat AI as license to compress thirty roles into three while simply piling on scope. Without that, you get the intensification the 2026 study found, not the economy of being.

4. Name the disorientation, don't paper over it. Senior leaders navigating this report not difficulty but disorientation: the old status markers dissolve. You used to matter because people reported to you. Now you matter because of what you see and what you choose. That's a more honest kind of importance — and it takes adjustment. Organizations that present the transition as pure liberation tend to lose the leaders they most need to keep.

The question of what humans are for has a stubbornly practical answer inside an AI-Born firm. If the agents are payroll, the humans are the thing payroll can't buy: the seat, the agenda, and the judgment that, once written down, keeps deciding after everyone's gone home. That's not a consolation. On a long enough horizon, it's the whole moat.