A practical classification framework for accounting, governing, and funding the workforce of humans and agents. Most enterprise finance and HR functions are running 2026 with a 2019 chart of accounts.
Boards will start asking by Q3 of next year. Most CFOs are a year away from being able to answer. Start now or start late.
Product-embedded AI is COGS. Productivity copilots are software opex. Labor-substitution agents are workforce capital. Most companies lump all three together. The third is the one that actually matters.
Agentic spend ÷ (employee + contractor + agentic spend). One number a board can track. Anything else hides the math.
Four decisions sit underneath every agent deployment. Default behavior is to dump all four on IT. That breaks every agent program at scale.
The narrow capex exceptions are real but limited to reusable internal infrastructure under ASC 350-40.
The chart of accounts, the metric, and the operating model are roughly four quarters of work. Boards will start asking by Q3 of next year.
Not all AI spend is the same. Treating it as one bucket is the source of the confusion. Three categories. Three places on the P&L.
If your platform uses an LLM to generate output for customers, that inference is part of the cost of delivering the product. Salesforce, Snowflake, ServiceNow now separate AI delivery cost in financial reporting.
ChatGPT Enterprise. Microsoft Copilot. GitHub Copilot. Glean. Note-taking AI. Vibe coding tools. The user is human. The output is for human consumption.
Klarna's customer service agent did the work of 700 reps. Agentforce reducing support headcount. Block's engineering productivity translating into a smaller hiring plan. The spend behaves like payroll, not software.
The test is simple because the economics are simple. Labor-substitution AI competes with hiring decisions, and belongs in the budget where hiring decisions are made. Try it on a real tool below.
Track it on the same line as the contractors and FTEs it substitutes for. Document the substitution math.
Once classification is right, the chart of accounts and the metrics need to follow. Start with one number a board can track.
One number that tells a board, investor, or regulator how much of your workforce capacity is digital. Try it with your own numbers.
Daylit's CTO Jerry Shu's practical test: if the AI is for operational efficiency, opex. If it's reusable infrastructure that enables future AI development, capex. Most agent spend fails the second test.
Cost of building them after the fact is roughly 5× the cost of designing the chart of accounts correctly now.
Every agent in production. Owner, function, business outcome, monthly spend, cost per outcome. Most companies don't have this list.
For agents making consequential decisions (credit, hiring, customer treatment): who approved, the policy framework, the human-in-the-loop design.
Tying agent spend to the benefiting function, with the substitution math (what would have been hired). Justifies workforce-line classification.
No single executive owns digital labor. The CIO can't. The CFO can't. The CHRO can't. Three executives have to operate as one decision unit, with clear lanes.
Most enterprises today have all four decisions land on the CIO by default. That's how agent deployments stall, lose budget visibility, and get classified wrong. The matrix above is the unlock.
Patterns we see across enterprises in the field today. None of them are subtle; all of them compound quarterly.
Burying labor-substitution agents inside a "SaaS — AI" line item. Hides the workforce conversation. Boards ask why margin didn't move; the answer is in a budget nobody's looking at.
The CIO procures the platform, deploys the agent, and ends up effectively making headcount decisions. The CHRO finds out from the org chart.
Trying to capitalize variable inference cost or per-outcome agent fees. Auditors will reverse it. Companies asking the capex question are usually trying to smooth a P&L line that should be visible.
"AI spend was $4M this quarter." Useless to a board. Without the three-category split, no one can tell whether margin is compressing, productivity is improving, or workforce is being recomposed.
Three quarters in, the company has thirty agents in production and no list. Cost per outcome unknown. Substitution math can't be defended to the board.
Deploying an agent without documenting which headcount it replaced or which hire it deferred. The agent never gets credit for the workforce impact; the budget conversation defaults to pure cost.
Five questions. Check the boxes you can honestly answer yes to today. If you score all five, you're ahead of the operating curve. Score none and the next twelve months matter.
AI spend split into product-embedded (COGS), productivity copilot (software opex), and labor-substitution (workforce capital). Visible to CFO, CHRO, and audit committee.
The metric reported quarterly. Trend visible over four quarters. Companion ROSE-plus calculated alongside.
Every agent in production listed with owner, business outcome, monthly spend, and cost per outcome. Auditable on demand.
CFO, CHRO, and CIO meet on a recurring cadence with function leaders to review spend by category, Mix trend, and the inventory. Decisions documented.
Workforce, budget, platform, and operations questions each land with their actual owner. No more default-to-IT.
The chart of accounts decides what gets governed. Workforce capital is changing faster than the chart.or read 2028's earnings calls without knowing what the numbers mean.Fix the chart in 2026
Three inputs, four benchmark bands, and a comparison against companies at your stage. The TeamOhana platform reports the metric natively.