Enterprise Software Watch: AI Agents Are Rewriting the SaaS Business Model — Not Just Adding a Feature

Enterprise Software Watch | September 17, 2026

The SaaS Business Model Is Being Rewritten by AI Agents

Executive Overview

AI Agents Are Not Another Feature. They Are Changing What Enterprises Buy.

For two decades, the enterprise software industry operated on a simple and extraordinarily profitable premise: charge per seat, grow revenue as headcount grows, and deliver software at near-zero marginal cost. That model built Salesforce, Microsoft, ServiceNow, SAP, Oracle, and Workday into some of the most valuable companies in the world. It is now under structural pressure.

Gartner estimates that up to $234 billion of enterprise application software spending could be exposed to agentic AI disruption through 2030, representing roughly 20% of SaaS application spending by then. The firm calls this "agentic arbitrage"—the process by which AI agents complete tasks across multiple systems, bypassing the user interfaces that enterprises currently pay to access. Gartner's George Brocklehurst put it bluntly: "Agentic systems deliver outcomes directly, bypassing traditional UX-heavy applications and making the software invisible. This breaks the link between user growth and revenue growth for many enterprise software vendors."

The Traditional SaaS Business Model

The SaaS model rests on four pillars: per-seat licensing, recurring subscriptions, expansion revenue, and high gross margins. Software is delivered over the internet, updated continuously, and priced per user per month. The economics are powerful because the marginal cost of serving an additional user is close to zero. Gross margins of 75–85% are standard. Net revenue retention above 110% is the benchmark of a healthy business.

This model worked because software was a tool for humans. More employees meant more seats. More seats meant more revenue. The application interface was the product—the place where work actually happened. For twenty-five years, Salesforce's per-seat subscription model defined the industry. It lowered upfront costs, shifted maintenance burdens to vendors, and created predictable recurring revenue.

AI agents change that. They are not users. They do not need a seat. And increasingly, they do not need the interface at all.

Why AI Agents Change the Economics

The core disruption is straightforward: one AI agent can perform work previously done by multiple human users. When an agent resolves a customer service ticket, processes a payroll transaction, or negotiates a contract, no human logs into the application. The seat-based revenue model loses its connection to value creation.

Simultaneously, AI introduces variable costs that SaaS never had. Inference, orchestration, retrieval, and observability consume compute resources. Gross margins compress. AI products averaged 45% gross margin in 2025 and are projected near 53% in 2026—against the 70–85% that SaaS built its valuations on. Inference alone accounts for roughly 23% of revenue at scaling-stage AI companies.

The result is a fundamental shift: software value increasingly depends on actions and outcomes rather than access and interfaces. As ServiceNow's president and chief product officer Amit Zavery observed, "Advisory AI has run its course; enterprises need AI that senses, decides, and securely acts in accordance with organizational guardrails."

Traditional SaaS vs. Agentic SaaS

Dimension	Traditional SaaS	Agentic SaaS
Pricing unit	Per seat / per user	Credits, actions, outcomes
Revenue driver	Headcount growth	Work volume/outcomes
Marginal cost	Near zero	Variable (inference, compute)
Gross margin	75–85%	45–60%
User interface	The product	Increasingly irrelevant
Retention risk	Moderate (switching costs)	High (agent substitution)

From Seats to Usage: The New Pricing Landscape

Every major enterprise software vendor has introduced a new pricing currency in 2026. SAP has AI Units. Microsoft has Copilot Credits and Agent 365 subscriptions. Workday has Flex Credits. Oracle has AI Units (priced at one cent each). ServiceNow has consumption-based Assist SKUs. GitHub Copilot has AI Credits. The result is a proliferation of vendor-specific metrics that make software procurement dramatically more complex.

As IT Pro reported in September 2026, "pricing has shifted towards vendor-specific units such as tokens, credits, work units, and currency multipliers, often layered on top of existing seat or subscription charges." One vendor's token is not another's. Credits get repriced. Enterprises must now model tokens, credits, actions, and variable AI consumption rather than simply counting seats.

Enterprise Software Pricing Models

Model	Billing Unit	Example	Risk
Seat-based	Human user	Salesforce $25–$500/user/mo	Agent substitution reduces seat count
Consumption	API calls, tokens, compute	Microsoft Copilot Credits ($0.01 each)	Unpredictable costs
Credit-based	Vendor-specific credits	SAP AI Units, Workday Flex Credits, Oracle AI Units	Opaque pricing, true-up risk
Transaction	Completed task	Salesforce $2 per autonomous resolution	Attribution disputes
Outcome-based	Business result	Revenue gain or cost savings	Measurement complexity

The New Enterprise Software Margin Problem

The traditional SaaS margin structure assumed near-zero marginal cost. AI destroys that assumption. Every agent action consumes compute. Every inference request costs money. Every retrieval operation, orchestration step, observability log, and security check adds to the cost of delivering software.

ICONIQ's January 2026 State of AI survey put AI-native product gross margins at 52% this year—up from 41% in 2024 but still 25–30 points below the 75–85% mature SaaS norm. Inference now eats roughly 23% of revenue at scaling-stage AI companies. AI products averaged 45% gross margin in 2025, projected near 53% in 2026.

The implication is stark: software companies cannot simply add AI features and expect to maintain SaaS-level margins. They must design for margin efficiency from the ground up—through model selection, caching strategies, retrieval optimization, and pricing structures that align cost with value.

AI Cost Structure

Cost Layer	Description	Margin Impact
Model inference	Per-token cost of running LLM queries	Largest variable cost; 20–40% of revenue at scale
Retrieval	Vector search, RAG, knowledge graph queries	Adds latency and compute overhead
Orchestration	Multi-agent coordination, tool calls, retries	Compounds inference costs
Observability	Logging, tracing, evaluation, compliance	Non-negotiable for enterprise deployment
Security & governance	Identity, permissions, audit trails, guardrails	Enterprise requirement; adds fixed cost
Support	Human escalation, exception handling	Hybrid human-AI model increases cost

Salesforce: Agentforce and the Post-Seat Model

Salesforce is making the most aggressive bet on outcome-based pricing. In August 2026, the company shifted Agentforce to a model where customers pay based on revenue gains or cost savings rather than seats. Agentforce annual recurring revenue passed $1.5 billion, up more than 240% year-on-year. CEO Marc Benioff framed the goal as charging $2 for every $20 or $40 in revenue the software helps generate.

The company has also launched Agentforce Help Agent at $2 per autonomous resolution—defined as a two-message interaction with non-negative feedback and no escalation. It acquired Fin for $3.6 billion to strengthen its AI customer service capabilities. And it launched Claudeforce, letting customers use Anthropic's Claude to complete tasks inside Salesforce apps, with plans to charge for every third-party AI call.

Why this matters: Benioff acknowledged the uncertainty directly, saying Salesforce is "following startups rather than leading the change." The company's per-seat model—which it pioneered 25 years ago—is cannibalizing itself. Salesforce shares surged 22.6% after its Q2 earnings, but the long-term question is whether outcome pricing can replace seat revenue at scale.

Microsoft: Copilot, Agents and Enterprise Bundling

Microsoft's strategy is bundling and consumption. Agent 365, generally available since May 2026, is priced at $15 per user per month standalone or included in the Microsoft 365 E7 suite at $99 per user per month. It provides a unified agent registry across Microsoft, AWS Bedrock, and Google Cloud—positioning Microsoft as the enterprise control plane for multi-cloud agent governance.

Copilot Credits are priced from $0.01 each, with consumption-based billing for agent actions. Copilot Cowork requires a $30 per user per month Copilot license plus usage-based charges. Microsoft is leveraging its hyperscaler position and M365 lock-in to capture multi-agent workflow budgets through enterprise agreements.

Why this matters: Microsoft is attempting to make agent orchestration a native capability of its productivity suite—bundled into enterprise agreements rather than sold as a standalone product. If successful, it makes the control plane the product, not the application.

ServiceNow: Monetizing Autonomous Workflows

ServiceNow's AI annual contract value crossed $1 billion in Q2 2026, with net new AI ACV growing more than 40% sequentially. Production deployments rose ninefold in nine months. More than 40 customers are running Level 1 IT service management AI specialists that resolve 80–85% of service requests without human interaction—requests that previously took two days now complete in about 20 minutes.

Critically, half of ServiceNow's net new business is now non-seat-based. New AI-native SKUs generate price uplifts in the 20–30% range. The company is targeting AI to represent over 30% of ACV by 2030 and expects to exceed $1.5 billion in AI ACV by year-end 2026.

Why this matters: ServiceNow is the clearest proof point that non-seat pricing can scale. Its AI Control Tower, Context Engine, and Action Fabric provide the governance and orchestration layer that enterprises need to deploy agents at scale.

SAP & Oracle: AI Inside Systems of Record

SAP has introduced AI Units as its consumption currency. Roughly 200 AI actions are bundled per Advanced Full User Equivalent, but a multi-step agent draws 5–10 times an interactive prompt—meaning the real allowance is 20–40 agent runs. Past the pool, an agent run costs $0.40 to $1.80. Since July 2026, usage-based pricing has become the default posture at cloud renewal, not an optional add-on.

Oracle has taken a similar approach with AI Units priced at one cent each, pooled across Fusion pillars, with 20,000 free units per month. Premium LLM usage consumes approximately 5 AI Units per action. Oracle's strategy is to make general actions free—the multiplier is literally 0x for basic LLM tasks—while charging for premium agentic workloads.

Why this matters: Both SAP and Oracle are betting that their systems of record—ERP, HCM, financials—remain essential, but the way enterprises consume and pay for those systems is changing. The meter is now on actions, not users.

Workday: AI Agents in HR and Finance

Workday's Flex Credits are a consumption model included in every subscription, metered per agent skill. The rate card runs from 1 credit for information retrieval to 750 credits for talent pool lead identification. A resume screen costs 6 credits. A contract review and redline costs 500 credits. Workday says its AI products generated more than $100 million in new annual contract value in Q2 2026, with more than 5,500 customers using agents.

Why this matters: Workday's model is designed to align cost with value—but it moves cost risk to the buyer. When the meter is per action, the buyer owns the volume. And volume is the one thing a vendor estimate cannot predict for a specific enterprise estate.

AI-Native Startups vs. Incumbent SaaS

The competitive dynamic between AI-native startups and incumbent SaaS vendors is more nuanced than a simple disruption narrative. Startups like Sierra, Fin, and Hang Ten Systems are attacking individual workflows with AI-native architectures, faster product development, and new user experiences. They have lower organizational complexity and the ability to rethink legacy workflows from scratch.

But incumbents retain structural advantages: existing customers, distribution, enterprise relationships, proprietary data, integration ecosystems, and existing contracts. SAP's Knowledge Graph gives its agents a structured map of business entities that startups cannot replicate. ServiceNow's agent platform operates directly on data and workflows already in ServiceNow.

The recent Cohere–Aleph Alpha merger is a useful example of enterprise AI vendors emphasizing regulated, deployable AI infrastructure. The combined company—valued at around $20 billion—will operate as Cohere with dual headquarters in Toronto and Berlin, focusing on sovereign AI deployments that run inside a customer's own computing environment. It's a bet that enterprises and governments want AI that is "powerful enough to compete, but secure and governable enough to trust."

The Enterprise Software Pricing Reset

Enterprise software procurement is becoming harder, not easier. AI add-ons, consumption pricing, overlapping tools, and complex licensing are making it difficult for enterprises to model costs. As IT Pro reported, pricing has shifted toward vendor-specific units—tokens, credits, work units, currency multipliers—often layered on top of existing seat or subscription charges.

Gartner predicts that at least 40% of enterprise SaaS spending will shift toward usage-, agent-, or outcome-based pricing by 2030, with seat-based vendor revenue share declining from 21% to 15%. But the transition is uneven. Only 4 of 65 enterprise software companies analyzed by AlixPartners have fully adopted outcome-based pricing. More than half still rely primarily on per-seat models.

Attribution disputes loom large. Payment processor Stripe has warned that sales conversions "may result from product changes, marketing campaigns, or seasonal factors" rather than the software itself. "Unless attribution rules are clear, customers may dispute whether results should be credited to the software vendor."

What Happens to SaaS Companies?

Four paths are emerging for incumbent SaaS companies:

AI-enhanced SaaS: Add AI features to existing products. Copilots, assistants, and embedded agents make applications easier to use but do not change the underlying business model. This is the least disruptive path—and the least defensible.
Agentic SaaS: Rebuild the product around AI agents that execute work autonomously. Pricing shifts from seats to actions or outcomes. ServiceNow and Salesforce are furthest along this path.
Platform consolidation: Become the orchestration layer that controls agents across multiple systems. Microsoft Agent 365, SAP AI Agent Hub, and ServiceNow AI Control Tower are competing for this position.
AI-native replacement: New entrants replace legacy applications entirely with agent-first architectures. This is the path most threatening to incumbents—and the hardest to execute at enterprise scale.

Most incumbents are pursuing a combination of the first three. The fourth remains the domain of startups and focused challengers.

The New Enterprise Software Stack

The enterprise software value chain is being restructured. The traditional stack—application → user interface → human action—is giving way to a longer, more complex chain:

Systems of Record → Enterprise Data → AI Models → Agents → Orchestration → Governance → Actions → Outcomes

Value is shifting away from the application layer toward the layers that control data, agents, and outcomes. The companies that own the systems of record—SAP, Oracle, Workday—retain an advantage in data and process logic. The companies that control agent orchestration—ServiceNow, Microsoft, Salesforce—are competing to become the new control plane. And the companies that govern agent behavior—identity, security, compliance—are becoming essential infrastructure.

The question is no longer whether AI changes enterprise software. It is which layer of the stack captures the economic value.

Strategic Questions for Enterprise Software Companies

What should an AI agent cost? Current pricing ranges from $0.01 per credit to $2 per resolution to 750 Flex Credits per talent identification run. There is no standard. The market is still discovering what an action is worth.
Who owns the customer relationship? If an agent from ServiceNow or Microsoft orchestrates work across Salesforce and SAP, which vendor owns the customer? The orchestration layer controls the relationship.
How much software can one agent replace? A single agent resolving tickets autonomously replaces multiple human users and the seats they occupied. The revenue implications are significant.
Can vendors preserve margins as AI usage increases? Higher usage means higher inference costs. Unless pricing scales with consumption, margins compress further.
Will enterprises consolidate around fewer platforms? If orchestration becomes the control plane, enterprises may standardize on fewer vendors—reducing the number of applications they buy.
Does the application remain the product if agents perform the work? When the interface ceases to be the primary point of interaction, what exactly is the customer buying?

Conclusion

The CODEW Analysis

AI agents are not simply another feature inside enterprise software. They are changing what enterprises buy, how vendors charge, and where software companies capture economic value.

Gartner's $234 billion warning is not a prediction of SaaS's death. It is a prediction of SaaS's restructuring. Seat-based revenue will decline. Consumption and outcome-based models will grow. Gross margins will compress. The companies that thrive will be those that design for the new economics—not those that try to preserve the old ones.

The key question for every enterprise software company is no longer "How do we add AI?" It is "Where in the value chain do we capture economic value when agents perform the work?" The companies that answer that question correctly will define the next decade of enterprise software. The companies that don't will watch their revenue model erode one seat at a time.

Incumbent Enterprise Platforms: AI Strategy and Monetization

Company	Core Platform	AI Strategy	Monetization
Salesforce	CRM	Agentforce, multi-model	$2/resolution, outcome-based
Microsoft	M365, Azure	Agent 365, Copilot	$15/user/mo + credits
ServiceNow	Workflow platform	Autonomous Workforce	Non-seat ACV, consumption
SAP	ERP	Joule agents	AI Units (actions)
Oracle	Fusion ERP/HCM	Fusion Agentic Apps	AI Units ($0.01 each)
Workday	HCM, Finance	Illuminate, Sana	Flex Credits (1–750/action)

Sources

Gartner — $234 Billion in Enterprise Application Software Spend Is at Risk from Agentic AI (July 1, 2026)
IT Brief Australia — Gartner warns agentic AI threatens $234bn SaaS spend (July 2, 2026)
Edgen — Salesforce bets on outcome pricing as AI upends 25-year SaaS model (Aug 31, 2026)
EPC Group — Microsoft Agent 365 GA: Registry Sync with AWS Bedrock + Google Cloud (May 2026)
Edgen — ServiceNow AI ACV tops $1B as agentic deployments surge ninefold (Aug 27, 2026)
Redress Compliance — SAP AI Units 2026: 200 Actions Buys 20 Agent Runs (Aug 16, 2026)
Redress Compliance — Workday Flex Credits: 2026 Pricing Guide (Aug 6, 2026)
Redress Compliance — Oracle Fusion AI Agents: The Meter (July 1, 2026)
IT Pro — Why enterprise software is becoming harder – not easier – to buy (Sept 15, 2026)
CloudZero — AI gross margin: how AI spend hits SaaS profitability (Aug 31, 2026)
ICONIQ — State of AI Survey (January 2026)
Reuters — Cohere, Aleph Alpha combine to target enterprise AI market (Sept 16, 2026)
AlixPartners — Outcome-based software pricing: Hype or reality? (July 29, 2026)
Futurum Group — Are Outcome-Based and Hybrid AI Pricing Models Rewriting the Vendor Playbook? (May 12, 2026)
ZDNet — The great software pricing shakeout of 2026 (Dec 24, 2025)

The CODEW Stat

$234 billion of enterprise application software spending is at risk from agentic AI through 2030. AI products averaged 45% gross margin in 2025, projected near 53% in 2026—against the 70–85% that SaaS built its valuations on. Half of ServiceNow's net new business is now non-seat-based. And 97% of SaaS CEOs plan to retire seat-based pricing within two years, yet only 4 of 65 enterprise software companies have fully adopted outcome-based pricing.

The CODEW