Your new headcount plan is a compute plan

I looked at a 2026 plan last week that had two spreadsheets. One was "Headcount." The other was "Infrastructure." They were owned by two different teams, reviewed on two different cadences, and approved by two different committees.

This is insane.

The unit economics have collapsed

In 2018, a dollar of headcount bought you roughly 1.0 dollar of headcount output. A dollar of compute bought you approximately zero dollars of headcount-equivalent output. They were not substitutes.

In 2026, they are substitutes. Not perfectly, not in every function, but meaningfully and measurably. Legal review, financial analysis, copywriting, first-line support, code review, even parts of IC engineering — the substitution ratio is no longer zero.

A CFO who plans these separately is pricing a blended commodity in two different currencies.

The new plan

Build one roster. Agents and humans on the same ledger. For each line:

• Cost per unit of output (fully loaded, not just wage)
• Latency (how fast does work come back)
• Floor and ceiling (what's the worst and best case quality)
• Substitutability (which other lines on the roster could do this work)

Suddenly capacity planning is legible. You can see where an agent beats a human on cost but loses on quality ceiling. You can see where a human is irreplaceable and invest there. You can see which roles are genuinely hybrid.

One closing thought

If your org chart is a Notion page and your agent inventory is a Jira board, you do not have an org chart. You have a half org chart. Draw the rest of it.