The New Workforce / Field Notes
A Field Brief · FN-01 · Workforce Capital · Category Formation

Half your headcount
doesn't have apulse

A thesis on the largest enterprise category since cloud. The category isn't AI software. It's workforce capital. The trillion-dollar AI spend forecast does not fit inside the IT budget. It fits inside the workforce budget. That changes everything.

Document
FN-01 · Field Brief
Audience
Investors & operators
Read time
9 minutes
Author
Tushar Makhija
CEO, TeamOhana
Version
1.0 · April 2026

The framing, said out loud

"Agents are virtual knowledge workers that can perform many tasks people do today. When you consider the global economy is $110 trillion, with nearly

$50 trillion tied to white-collar knowledge workers
, you begin to see the scale of market potential: agents can disrupt or enhance a total addressable market
bigger than all other industries combined
."

Joe Tsai
Chairman, Alibaba Group
Siemens RXD Summit · Beijing · March 2026

Four numbers

The numbers
that broke the model.

Two of the most exposed CEOs in enterprise tech, US and China, have converged on the same framing: the AI agent TAM cannot be measured against software. It is sized against the labor pool itself.

0
T
White-collar wage bill the agent TAM is being sized against
Joe Tsai · Alibaba · Mar 2026
$
0
T
Cumulative digital labor spend by 2030
IDC · Oct 2025
$
0
T
Worldwide AI spend in 2026, +44% YoY
Gartner · Jan 2026
0
%
CAGR on AI tech supporting digital labor
IDC · Oct 2025

Joe Tsai sized the AI agent market last month at $50 trillion, against the global white-collar wage bill — not against the enterprise software budget. He is right about the number and right about the math.

Every category model in enterprise software just became obsolete.

§ 01

The number that
breaks the model.

For thirty years, every enterprise software category was sized against software budgets. That math no longer works.

CRM was a $200B market because that's what companies spent on customer relationship tools. ERP was sized against finance and operations software. Even the cloud was sized against the on-prem infrastructure it replaced.

The fastest-growing category in enterprise technology is not being funded out of software budgets.

It is being funded out of payroll.

1.1 — The math no longer closes

IDC's October 2025 Digital Labor Economy report puts cumulative spending on AI technologies supporting digital labor at

$3.34 trillion by 2030
, growing at a 43% CAGR. Gartner's January 2026 forecast puts worldwide AI spending at
$2.52 trillion in 2026 alone
, up 44% year-over-year. For reference, the entire worldwide IT spend in 2026 is forecast at $6.15 trillion. AI is now roughly 41% of total IT spend.

Three budgets · 2026, in trillions

The AI category is too big to fit inside software. It only fits inside labor.

IT budget
AI / digital labor
White-collar wages
Enterprise software
The market AI used to size against
$0.65T
2026
Worldwide IT spend
Where AI gets booked today
$6.15T
2026 · Gartner
AI spend, 2026
41% of all IT spend
$2.52T
+44% YoY
White-collar wage bill
The new sizing base
$50T
Tsai · 2026

The global enterprise software market is roughly $650B. The numbers above are sized against something far larger — the labor pool itself. Marc Benioff put the digital labor TAM at "$3 trillion to $12 trillion" and called it bigger than software. Tsai's $50T number sizes the same TAM against the white-collar wage bill, just from the supply side rather than the vendor side.

Redpoint Ventures' 2026 CIO survey adds the buyer-side proof: task agents alone expand the addressable market

from $0.5T to $1.2T
. The incremental $700B has to come from somewhere other than software. There is only one budget large enough to fund it.

§ 02

What CFOs
are actually doing.

Stated strategy is theater. Capital allocation is signal. Look at what CFOs are actually doing.

Gartner's February 2026 CFO budget benchmarks (n=303) show headcount growth expectations collapsing from 6% in 2025 to 2% in 2026. Only 21% of CFOs plan staff increases of 4 to 9%, down from 31% the year prior. Gartner's analyst Nauman Abbasi calls this

"a structural pivot from labor expansion to optimization driven by automation and AI."

Sixty percent of CFOs plan to increase finance function AI investments by 10% or more. HR budgets saw the sharpest pullback in the entire benchmark. The cause Gartner cites:

"reduced hiring and AI efficiency gains."

This is not a story about technology adoption. It is a story about budget reallocation.
Money that was going to humans is now going to agents.

The capital reallocation, in public

Four data points from the last six months. Each one a public number. Each one funded out of payroll, not IT.

$
50
T
White-collar TAM cited by Alibaba's chairman as the agent sizing base.
Joe Tsai · March 2026
6%
→ 2%
CFO headcount growth expectation, year-over-year. The largest single-year collapse Gartner has recorded.
Gartner CFO Benchmarks · Feb 2026
853
FTE-equivalents run by Klarna's AI customer-service agent, disclosed on Q3 FY2026 earnings.
Klarna Earnings · Q3 FY2026
$
540
M
Agentforce ARR, +330% YoY. The fastest product ramp in Salesforce's history.
Salesforce · Q3 FY2026

2.1 — Codified in policy

Shopify codified it in policy. Tobi Lütke's April 2025 memo:

"Before asking for more headcount and resources, teams must demonstrate why they cannot get what they want done using AI."
Shopify headcount has fallen 35% cumulatively since 2022 even as AI spend has climbed.

Block cut from 10,000 to under 6,000 in February 2026. Jack Dorsey's shareholder letter:

"intelligence tools have changed what it means to build and run a company."
Citigroup's CFO is targeting a 20,000 headcount reduction by 2026 with AI cited as the lever.

The cash funding agentic AI is being pulled out of payroll.

The accounting category may still say IT. The economics say labor.

§ 03

The pricing model
gives the game away.

Software is priced per seat. Labor is priced per outcome. AI agent pricing is moving to outcomes.

Era one · The SaaS decade

Software, priced per seat.

$

/seat/mo

  • Unit
    A human user logged in.
  • Margin
    85% gross margin. Cost is distribution, not delivery.
  • Budget
    IT opex. Approved as software.
  • Buyer
    CIO. Procurement-led.
  • Example
    Zendesk at $115 per support agent seat.

Era two · The agent decade

Labor, priced per outcome.

$

/resolution

  • Unit
    A ticket resolved, a deal advanced, a task completed.
  • Margin
    30 to 60%. Inference cost is COGS, not overhead.
  • Budget
    Labor envelope. Approved as workforce capital.
  • Buyer
    CFO + CHRO. Function-led.
  • Example
    Decagon, priced per conversation resolved.

3.1 — The Salesforce signal-inside-the-signal

Salesforce launched Agentforce at $2 per conversation in late 2024. It pivoted in May 2025 to Flex Credits at $0.10 per action. It added per-user licenses at $125/month in late 2025.

Three pricing models running simultaneously
, generating $540M ARR by Q3 FY2026 with 330% YoY growth.

The signal inside the signal is the Flex Agreement. Salesforce customers can

convert human user licenses into Flex Credits and back.
You cannot convert your AWS bill into your HR seats. You can convert your Salesforce seats into Agentforce credits. That fungibility is the budget-category collapse made operational. It is the most consequential pricing decision in enterprise software since per-seat replaced per-server, and the second-order effects on how Finance models the rest of the SaaS stack are going to take three to four years to fully play out.

3.2 — Not US-only

Alibaba spun out the Alibaba Token Hub as a standalone business group in March 2026, restructuring the entire commercial model around

tokens
— units of generated and actioned intelligence, rather than compute resources. Citi Research called it the shift from
"selling resources to selling intelligent capabilities."

When the largest cloud platform outside the US reorganizes its P&L around action rather than infrastructure, the repricing is structural, not a SaaS-vendor experiment.

Software pricing is fragmenting because software is no longer the unit being sold.

§ 04

The displacement economics
are now public.

When the CFO modeled this investment, they did not compare "Agentforce price per user" to "Zendesk price per user." They compared AI assistant cost to avoided customer service agent salaries.

Klarna is the cleanest publicly audited case. The Q3 FY2026 earnings call disclosed AI customer service handling the equivalent work of 853 FTEs, $60M in cost savings, two-thirds of all customer inquiries automated, resolution time down 82%. Total deployment cost was $2 to $3M. The payback period was under one quarter.

When the unit of comparison is the FTE, not the seat, the payback math

does not look like software.
It looks like a hiring decision that paid back in one quarter.

853
FTE-equivalents replaced by the AI customer service agent
$60M
Annual cost savings, audited and disclosed on the Q3 call
82%
Reduction in resolution time per ticket
<1Q
Payback period on a $2 to $3M deployment

Gartner forecasts

$80B in contact center labor cost reductions by 2026
from conversational AI. The line item in Gartner's report is "agent labor costs," not "IT cost growth."

A large financial services firm in Accenture's March 2026 banking research has made its AI agents formal

"digital employees"
— with logins, email accounts, system access, and human managers. Accenture reports the emergence of "agent descriptions" as parallels to job descriptions.

"In a lot of ways, the IT department of every company is going to be the HR department of AI agents in the future."
Jensen Huang · Deloitte research · December 2025

Agents get evaluated like employees, governed like employees, provisioned like employees.

The unit of value is the FTE, not the seat.

§ 05

Why this matters for
category formation.

Every durable enterprise category has been built on a redefinition of the budget it served. Workforce Intelligence has the same shape.

Salesforce did not win because it was better Siebel. It won because it created the SaaS budget category and got procurement to accept that operating expense was fundable in a way that capital expense was not. Workday won by collapsing HRIS, payroll, and finance into one ledger. Snowflake won by separating compute and storage. ServiceNow won by becoming the system of record for the IT service desk.

The Workforce Intelligence category has the same shape.
It is not a software category. It is a budget category. It is the budget category that governs how organizations allocate, plan, and account for the blended workforce of humans and agents.

5.1 — The read path and the write path

The read path
Workday lives here.
Workday records what happened — who you hired, who left, who was paid what. It is the system of record after the fact.
The write path
This is where Workforce Intelligence sits.
The decisions about what to fund, who to hire, and how to allocate workforce capital all happen
upstream
of any system of record. Today they happen in spreadsheets and Slack threads, and get recorded after the fact. That gap is what the new category fills.

The companies that will compound from here are the ones that govern workforce capital allocation

before
the decision reaches any system of record. Every workforce decision is a capital allocation decision: hire a human, deploy an agent, retain a contractor, expand a function, freeze a team. The infrastructure to govern those decisions does not exist in Workday, in NetSuite, or in any HRIS or FP&A tool today.

The supply side is already organizing around this reframe. Salesforce Flex Credits convert human seats into agent capacity. Alibaba's Token Hub repackages cloud as actioned intelligence. Decagon prices per resolution.

The hyperscalers and the AI-native vendors are not selling software with AI features — they are selling outputs that substitute for labor, priced and provisioned accordingly.

The governance layer is conceding the same point. Zylo, the Gartner Magic Quadrant leader for SaaS Management with $75B of SaaS spend in their dataset, named

"AI Becomes the Most Expensive Invisible Worker"
as their #1 prediction for 2026, and their public position is that AI usage must be tracked like labor, not like software. Zylo is the category leader of the layer that was supposed to catch this spend, openly acknowledging the spend is leaving their category for a category that does not exist yet.

§ 06

The bear case,
and why it doesn't hold.

Three objections worth taking seriously. None of them close the math.

Bear · 01
"
AI spend is just replacing other software spend.
"

Redpoint's CIO survey shows 45% of AI budgets are replacing existing software. That is real.

But the math does not close.
Zylo's 2026 SaaS Management Index, drawn from $75B of actual SaaS spend across 75,000 customers, shows AI-native app spend up 108% year-over-year, with growth as high as 400% at large enterprises. That is not budget rotation inside the SaaS category. That is new dollars entering the category, faster than any line item in software has ever grown.

If AI were only replacing software, the addressable market would compress, not expand. Instead it is expanding by trillions. The remaining 55% has to come from somewhere.

Labor and services budgets are the only candidates with the scale to absorb the spend.

The supply side confirms it. Alibaba did not spin out a Cloud Software business group in March 2026. It spun out a Token Hub. Salesforce did not launch a new SKU pricing model. It launched a fungibility mechanism between human seats and agent credits.

Klarna walked back the framing of full replacement.

They did not walk back the spend.
They are still running 853 FTE-equivalents through AI, saving $60M annually, and projecting headcount reduction from 3,000 to 2,500.

The lesson from Klarna is not that AI cannot replace labor. It is that pure replacement underperforms blended deployment. The blended deployment still funds itself out of labor budgets, with a more sophisticated capital allocation model.

Correct,

and that is exactly the bull case.
Copilots are productivity software. They live in IT opex and they will continue to. Autonomous agents are the new category.

Gartner forecasts task-specific AI agents in enterprise applications will go from less than 5% of apps in 2025 to 40% by end of 2026.

That growth curve is not coming out of the productivity software budget. It is coming out of the payroll line.

§ 07

What this means for
the next decade.

A few predictions worth committing to in print.

01

Accounting · 3 years

The chart of accounts will change.

Within three years, the standard enterprise general ledger will have a distinct sub-account for digital labor that sits

inside the benefiting function
— Sales, Service, R&D, Finance — rather than inside IT. The SaaS CFO playbook is already published. The Big 4 advisory practices are already selling the rebuild. Workday is already shipping the Agent System of Record.

02

Buying · 2 years

The buyer will shift.

AI agent procurement will move from the CIO to a

joint motion between CFO and CHRO
. The CIO's role becomes platform and infrastructure governance. The decision about which agent to deploy, where, and at what budget will be a workforce capital allocation decision — owned by the people who own workforce capital.

03

Infrastructure · 5 years

A new system of record emerges.

Workday is the system of record for

who you hired
. There is no system of record for
why
you hired them, what you considered, what you funded instead, or what you decided not to do. That decision layer — the Workforce Decision Record — is the missing infrastructure. The companies that build it will compound the way Workday compounded against the legacy HRIS market.

§ 08 · The category bet

Half of every company's headcount in 2030 will not have a pulse. The agents will be in the org chart.
They will be in the budget. They will be in the workforce plan.

The infrastructure to govern that workforce

does not exist yet
. The category is being formed in real time. The TAM is not measured against software. It is measured against the global wage bill — roughly $40 trillion annually.

Even a small percentage of that, governed through a decision and intelligence layer, builds a generational company. This is the largest enterprise category formation event since cloud.

The rest will be sized against the wrong market.

Agent Labor Governance

Join the
waitlist.

We're building the governance layer for agent labor —

registry, FinOps gate, Companion Model, Digital Labor Mix reporting
. Get early access, design-partner invites, and the next iteration of this playbook.

Work email
Request access

We'll reach out when a slot opens for your org.

"

Agents are virtual knowledge workers that can perform many tasks people do today.

Joe Tsai · Alibaba · March 2026

"

AI usage must be tracked like labor. AI-powered SaaS costs are dynamic, distributed, and often invisible until it is too late.

Zylo · 2026 SaaS Management Index

FN
02
Coming next

The CFO and CHRO playbook for digital labor.

The investor thesis becomes operating reality through accounting, governance, and capital allocation. Read the practical classification framework for the workforce of humans and agents.

Read FN-02 →

Sources & citations

Primary sources only.

  1. Joe Tsai, Siemens RXD Summit (March 2026)
    The $50T white-collar wage bill framing. Beijing keynote.
    alibabagroup.com
  2. IDC, "Worldwide Digital Labor Economy Report" (October 2025)
    $3.34T cumulative digital labor spend by 2030. 43% CAGR.
    idc.com
  3. Gartner, "Worldwide AI Spending Guide" (January 2026)
    $2.52T worldwide AI spend in 2026, +44% YoY.
    gartner.com
  4. Gartner CFO Budget Benchmarks, Feb 2026 (n=303)
    The 6% → 2% headcount growth collapse. Nauman Abbasi commentary.
    gartner.com/finance
  5. Klarna Q3 FY2026 Earnings Call
    853 FTE-equivalents, $60M savings, 82% resolution-time reduction.
    klarna.com/press
  6. Salesforce, Agentforce / Flex Credits (May 2025 → Q3 FY2026)
    Three-model pricing. $540M ARR. The Flex Agreement seat-to-credit fungibility.
    salesforce.com/news
  7. Andreessen Horowitz, "Software is Becoming Labor" (Dec 2024 + Mar 2026 follow-up)
    The thesis and the per-seat-is-no-longer-atomic framing.
    a16z.com
  8. Alibaba, Token Hub spin-out (March 2026)
    The reorganization around actioned intelligence. Citi Research commentary on "selling intelligent capabilities."
    alibabagroup.com
  9. Redpoint Ventures, 2026 CIO Survey (March 2026)
    Task agents expanding TAM from $0.5T to $1.2T. The 45% replacement figure.
    redpoint.com/2026-market-update
  10. Zylo, 2026 SaaS Management Index
    $75B SaaS spend dataset. AI-native app spend +108% YoY (up to 400% at large enterprises).
    zylo.com/research
  11. Shopify, Tobi Lütke memo (April 2025)
    The "demonstrate why you cannot use AI" hiring policy.
    x.com/tobi
  12. Block, Q4 2025 shareholder letter (February 2026)
    Dorsey's headcount reduction from 10,000 to under 6,000.
    investors.block.xyz
  13. Jensen Huang in Deloitte, "AI & the Future of Work" (Dec 2025)
    "The IT department of every company is going to be the HR department of AI agents."
    deloitte.com
  14. Accenture Banking Research, March 2026
    "Digital employees" with logins, email, system access. "Agent descriptions" emergence.
    accenture.com

"The category isn't AI software.

It's workforce capital.
The trillion-dollar AI spend forecast does not fit inside the IT budget — it fits inside the workforce budget. That changes everything."

TeamOhana · Workforce Intelligence for the Age of Digital Labor
FN-01 · April 2026
↑