Is the product budget now a compute budget?

THE SHORT ANSWER

Increasingly, yes. In an agent-native company a growing share of product output comes from compute (tokens, tool calls, GPU time) rather than added headcount, so the cost shifts from fixed salaries to variable, usage-driven spend. A team of six running a fleet of agents produces what twenty used to. Three things change for the CPO and CFO: you budget output not heads, marginal cost stops being zero so waste becomes financial, and capacity becomes elastic. The org that plans in this blended unit wins on capital efficiency.

Every annual planning cycle I've sat through ran the same way. The product org submitted a headcount plan, more output meant more people, and cost grew in steps you could predict. The CFO drew a line a year out and trusted it. All of it rested on one assumption: output scales with heads. That assumption is breaking, and the planning stacked on top of it hasn't caught up.

How product used to be budgeted

The old model was clean. Headcount was the unit of capacity. You sized the team to the roadmap, the cost was salaries plus overhead, and it grew in steps as you hired. Fixed, in the accounting sense. A quiet week and a launch week cost the same, and that predictability was the whole appeal. Finance could plan, the board could model burn, the CPO could defend a number.

Then output stopped being a pure function of headcount. A team of six running a fleet of agents now produces what twenty used to. I've watched it happen and the math isn't subtle. A real chunk of the capacity comes from compute, and compute shows up as tokens, tool calls, and GPU time, not salaries. Opposite cost shape too. It's variable, it scales with usage instead of the org chart. Blend a fixed headcount line with a variable compute line and the combined product cost stops looking like payroll and starts looking like an infrastructure bill.

Three things change when you budget like this

You budget output, not heads. The question moves from how many people do we need to how much outcome do we need, and what does it cost to make. Cost per outcome replaces cost per FTE. Uncomfortable, because it's harder to forecast and easier to be held to. The discomfort is the job now.

Marginal cost stops being zero. Under the old model, once the team was hired, an extra unit of work was free. Now every agent run has a price. A badly scoped agent burning tokens on low-value work becomes a line item. Quality and efficiency stop being engineering concerns and become budget concerns.

And capacity gets elastic. You could never scale a human team up for a launch and back down the week after. Compute you can. The orgs that exploit this run lean baselines and surge when it counts.

So instrument compute spend at the workflow level, then bring finance one blended picture: the fixed headcount line, the variable compute line, and cost per outcome as the number that ties them together. Here's the one figure to pull this week. Your total agent and compute spend for the last quarter, divided by a unit of output you care about. Most CPOs have never calculated it.

SOURCES

THE LONG VERSION

RELATED ANSWERS

Last reviewed 2026-07-31 · 3 min read