Industry View · Enterprise Software

The seat is dying. The agent is the new line item.

Enterprise software is being repriced in real time. Global business-software spend is set to grow 14.7% to more than $1.4 trillion in 2026, yet the application layer just lived through a $2 trillion drawdown as investors bet that autonomous agents erode the per-seat model that built SaaS. The fight is no longer cloud-versus-on-prem; it is whether incumbents can sell the agent before a startup sells it for them.

Demonstrated ROIROI Classification

Agentforce passed $1B run rate; Now Assist ~$600M ACV; agent revenue monetized

Key Figures

$1.4T
2026 business software spend (+14.7%)
Gartner
40%
enterprise apps with task-specific AI agents by end-2026 (from <5% in 2025)
Gartner
$2T
software market-cap drawdown, late 2025-early 2026
Fortune / SaaStr
20M
Microsoft 365 Copilot paid seats (FQ3 2026)
Microsoft / No Jitter

Value Chain

Foundation models
The horizontal substrate

Frontier LLMs from OpenAI, Anthropic and Google are now the reasoning engine inside every enterprise agent, turning model access into a cost-of-goods line for software vendors.

OpenAI, Anthropic, Google
Agent platforms
Where incumbents re-meter

System-of-record vendors wrap models in governed, data-grounded agent runtimes and charge by consumption or outcome rather than by seat.

Salesforce Agentforce, ServiceNow Now Assist, SAP Joule
Horizontal copilots
Productivity at the OS layer

Copilots embedded in the daily work surface monetize the installed base but face the sharpest seat-cannibalization risk as agents absorb the tasks.

Microsoft 365 Copilot, Glean
Vertical agents
Outcome-priced point solutions

Startups attack single high-volume workflows — support, coding, finance ops — and price per resolution, directly displacing seats and BPO labor.

Sierra, Decagon, Cursor
Build vs buy
The enterprise re-evaluates the stack

Agent tooling lowers the cost of replacing thin SaaS with internal builds, pressuring vendors whose value was UI convenience rather than data or distribution.

Klarna-style in-housing, a16z portfolio

01 · The thesis

The unbundling of the seat

For two decades enterprise software was priced per user per month. That model assumed software was a tool a human operated. Agentic AI breaks the assumption: when an agent does the work, the seat it occupied disappears, and revenue tied to headcount shrinks with it. The market has already started pricing this. Application software now trades at a discount to the S&P 500 for the first time on record, and seat-heavy names with weak data moats have been punished hardest — Figma down ~86%, Duolingo ~83%, Monday.com ~80% from their highs. The incumbents' answer is to re-meter. Salesforce, ServiceNow, SAP and Microsoft are racing to sell the agent itself on consumption or per-outcome pricing — Salesforce's Agentforce passed a $1 billion run rate, and ServiceNow's Now Assist exited 2025 around $600 million ACV and is tracking toward a $1 billion-plus target for 2026, both monetized outside the seat. The winners will be those who own proprietary workflow data and a system of record agents must call; the losers are thin UI layers whose only moat was being where the human logged in.

The horizontal substrate

Frontier LLMs from OpenAI, Anthropic and Google are now the reasoning engine inside every enterprise agent, turning model access into a cost-of-goods line for software vendors.

Where incumbents re-meter

System-of-record vendors wrap models in governed, data-grounded agent runtimes and charge by consumption or outcome rather than by seat.

Productivity at the OS layer

Copilots embedded in the daily work surface monetize the installed base but face the sharpest seat-cannibalization risk as agents absorb the tasks.

Outcome-priced point solutions

Startups attack single high-volume workflows — support, coding, finance ops — and price per resolution, directly displacing seats and BPO labor.

The enterprise re-evaluates the stack

Agent tooling lowers the cost of replacing thin SaaS with internal builds, pressuring vendors whose value was UI convenience rather than data or distribution.

02 · The two clocks

Three clocks running against the per-seat model

Adoption clock. Gartner projects that 40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from fewer than 5% in 2025 — an eightfold shift in a single year that resets how every vendor must package and price its product. Spend clock. Money is still flowing in: business-software spend grows 14.7% to over $1.4 trillion in 2026, with AI accounting for a rising share of IT budgets. The growth is real, but it is concentrating toward AI applications and price increases rather than incremental seats. Repricing clock. The market has moved faster than the P&L. Roughly $2 trillion in software market value evaporated from the late-2025 peak into early 2026, and application software now trades at a discount to the S&P 500 for the first time on record (SaaStr / Fortune).

03 · Public players & exposure

Who routes through, who gets routed around

We plot the listed players on two editorial axes — how exposed each is to AI disruption, against how ready its data, brand and position are to be the answer. The figures in the table are sourced; the placement is our read.

04 · Private flagships

The AI-native challengers

The companies attacking this industry AI-first, with disclosed funding where available:

Salesforce Agentforce

The most aggressive incumbent pivot: consumption-priced agents sold on top of the CRM system of record, aiming to charge for the agent rather than lose the seat.

ServiceNow Now Assist

Agentic add-ons sold as consumption-based Assist packs across IT, HR and customer workflows, leveraging ServiceNow's position as the enterprise control tower.

Anysphere (Cursor)

AI code editor that became one of the fastest B2B software companies to scale past $1B ARR, redefining developer tooling as agent-assisted by default.

Sierra

Outcome-oriented AI agents for customer interactions, founded by ex-Salesforce co-CEO Bret Taylor, attacking the support seat and BPO spend directly.

Glean

Horizontal search-and-agent platform over enterprise knowledge, positioning as the connective tissue agents need across systems.

Decagon

Customer-support agents priced per-resolution — the clearest example of outcome pricing replacing the seat with a paid result.

05 · Signals

What moved, and what to watch

06 · The exposure read

Who’s defensible, who’s at risk

AI rewards clean, structured advantage and punishes friction. The line runs through who owns the data, the brand and the customer — and who is merely a step the technology can route around.

Sources

Where this comes from

The spend, and the payoff

The seat-economics drawdown: selected SaaS names off 52-week highs

Peak-to-trough decline during the 2026 SaaS rout. Source: 24/7 Wall St, Apr 2026. (% down from high)

Who's defensible, who's at risk

Defensible vs At Risk

Defensible

  • Systems of record with proprietary data — Salesforce, ServiceNow and SAP own the workflow data and governance agents must call, letting them re-meter on consumption instead of losing seats.
  • Distribution-anchored incumbents — Microsoft converts its installed base into 20 million paid Copilot seats, monetizing AI where users already work rather than fighting for net-new logins.
  • Outcome-priced agent startups — Sierra, Decagon and Cursor sell results (resolutions, completed code) and grow ARR faster than almost any prior software cohort by capturing labor budgets, not software budgets.
  • AI+human trust layers — Intuit's verified agents show that pairing autonomy with human checks drives repeat usage and willingness to pay more, a defensible position in regulated, high-stakes workflows.

At Risk

  • Thin seat-based SaaS — products whose moat was a convenient UI on top of someone else's data are pricing in 80%+ drawdowns; Figma, Duolingo and Monday.com are the market's chosen examples.
  • Headcount-linked revenue — any vendor billing per user faces shrinkage as agents reduce the number of humans who need a login, the quiet killer behind the seat-economics collapse.
  • Undifferentiated point tools — single-feature apps are the first build-vs-buy casualties as agent tooling makes internal replacements cheap, echoing Klarna-style in-housing of parts of the stack.
  • Premium-multiple SaaS without an AI story — with software now trading at a discount to the S&P 500, names that cannot show consumption or outcome-based AI revenue lose the valuation premium they long enjoyed.

The signals — how it unfolded

Aug 2025

Gartner: agents go mainstream

Gartner predicts 40% of enterprise apps will feature task-specific AI agents by end-2026, up from under 5% in 2025.

Late 2025-early 2026

The SaaS rout

Roughly $2 trillion in software market cap evaporates from the peak; seat-based names like Figma, Duolingo and Monday.com fall 80%+ from highs.

Feb 2026

Software cheaper than the index

For the first time on record, application software trades at a discount to the S&P 500 as investors reprice seat risk.

Q1 2026

Incumbents push AI past $1B

Salesforce Agentforce passes a $1B run rate and ServiceNow's Now Assist (over $600M ACV exiting 2025) tracks toward a $1B-plus 2026 target — agents monetized outside the seat.

May 2026

Agent capital intensifies

Sierra raises $950M at $15.8B and SAP reports two-thirds of Q4 cloud deals carrying Business AI, confirming demand is structural.

Challengers to watch

Salesforce Agentforce

The most aggressive incumbent pivot: consumption-priced agents sold on top of the CRM system of record, aiming to charge for the agent rather than lose the seat.

ServiceNow Now Assist

Agentic add-ons sold as consumption-based Assist packs across IT, HR and customer workflows, leveraging ServiceNow's position as the enterprise control tower.

Anysphere (Cursor)

AI code editor that became one of the fastest B2B software companies to scale past $1B ARR, redefining developer tooling as agent-assisted by default.

Sierra

Outcome-oriented AI agents for customer interactions, founded by ex-Salesforce co-CEO Bret Taylor, attacking the support seat and BPO spend directly.

Glean

Horizontal search-and-agent platform over enterprise knowledge, positioning as the connective tissue agents need across systems.

Decagon

Customer-support agents priced per-resolution — the clearest example of outcome pricing replacing the seat with a paid result.

Exposure table

CompanyStanceThe sourced fact
Microsoft MSFTDistribution moatMicrosoft 365 Copilot reached 20 million paid seats by fiscal Q3 2026, up from 15 million the prior quarter (Microsoft, via No Jitter).
Salesforce CRMRe-metering the seatAgentforce surpassed a $1 billion annual run rate and served roughly 18,500 enterprise customers, up from 12,500 a quarter earlier (Salesforce / VentureBeat).
ServiceNow NOWWorkflow control towerNow Assist exited 2025 at over $600 million in ACV and is tracking toward a $1 billion-plus AI ACV target for 2026 (ServiceNow / CX Today).
SAP SAPData-of-record moatTwo-thirds of SAP's Q4 2025 cloud orders included Business AI, and total cloud backlog hit €77 billion, up 30% YoY (SAP / ASUG).
Intuit INTUAI+human verifiedMore than 3 million users interacted with Intuit's autonomous AI agents, with over 85% returning for repeat engagements (Intuit, FY Q2 2026).
Anysphere (Cursor) pvtFastest scalerCursor reached ~$1B+ ARR and raised $2.3B at a $29.3B valuation in Nov 2025, with later talk of a round near a $60B valuation (CNBC / The Next Web).
Sierra pvtSeat displacerBret Taylor's Sierra raised $950M at a $15.8B valuation, having topped $150M ARR in eight quarters (TechCrunch / CNBC).
Glean pvtWork-AI platformGlean raised $150M at a $7.2B valuation, up from $4.6B nine months earlier (Crunchbase News).
Decagon pvtPer-resolution pricingDecagon was valued at $4.5B after a $250M Series D and prices support agents per-conversation and per-resolution (CMSWire / Built In).
Figma FIGSeat-exposedFigma fell about 86.5% from its 52-week high during the 2026 SaaS rout, among the hardest-hit seat-based names (24/7 Wall St).

Sources