Enterprise Software Watch: SaaS Vendors Build Toll Booths for AI Agents
The AI Agent Becomes the New Enterprise Interface — SaaS Vendors Build Toll Booths for Agents
1. Executive Brief
On August 27, the enterprise software market is being repriced in real time. PYMNTS reports that ServiceNow, SAP, and Workday are drawing new lines around customer data stored inside their platforms as external AI agents begin to erode the per-seat pricing model that has defined enterprise software for two decades. As enterprise AI replaces predictable per-seat billing with consumption models that behave less like subscriptions and more like utility invoices, finance teams are being forced to manage spending that fluctuates with model activity rather than headcount.
Here is the number making CFOs nervous: a single AI agent can fire off 50,000 API calls in a month. Under the per-seat pricing model that governed enterprise software for decades, those 50,000 calls would cost nothing extra. Under the new models being built by ServiceNow, SAP, Workday, HubSpot, and Datadog, they could cost thousands of dollars. ServiceNow announced Action Fabric at Knowledge 2026, an integration layer that allows external agents to execute governed workflows on its platform. The company's COO confirmed that it will meter this activity, meaning customers will pay per action. Workday said that charging for agent access offered considerable financial upside. Datadog, meanwhile, capped third-party agents at 5,000 daily requests against its MCP server.
The CODEW thesis: The next phase of enterprise software will be defined not by how many applications employees use, but by how much business work AI agents execute across them—and who gets to charge for it. The toll booth, not the seat, becomes the new enterprise software business model.
2. Enterprise Software Market This Week
Enterprise SaaS has crossed into an AI displacement zone, with ServiceNow triggering a reported $500 billion repricing. Salesforce, Workday, and Oracle fell in lockstep, erasing more than $80 billion in market value as investors shifted their AI thesis from software enhanced by AI to software potentially replaced by AI agents. The synchronized selloff marks an inflection point in which enterprise application vendors are transitioning from AI beneficiaries to potential displacement candidates, forcing valuation resets across a sector worth more than $500 billion. The reported 16% drop represents a moment when Wall Street began pricing in the possibility that AI agents could undermine the enterprise SaaS revenue model before vendors complete their AI pivots. The contagion tells the larger story: when Salesforce, Workday, and Oracle all moved lower together, investors signaled that they viewed this as a sector-wide business-model question rather than a ServiceNow-specific execution issue.
The software industry is moving through three distinct pricing phases. Per-seat: a fixed price per human user per month, used by companies such as Salesforce, Workday, and Atlassian, is under existential pressure as AI agents reduce the number of human seats required. Usage-based pricing tied to consumption, such as API calls, credits, or storage, appears more resilient because AI agents increase consumption. Outcome-based: pricing tied to the business result delivered by software is emerging as another model. By successfully transitioning toward outcome-based pricing, ServiceNow is helping establish a new benchmark. The era of seat-count erosion fears—where investors worried AI would allow companies to shrink their workforces and therefore their software needs—is increasingly being replaced by an era in which providers are paid for the productivity AI agents deliver.
What changed: The tollgate strategy turns disruption into a new revenue line. Who gains: ServiceNow and Workday, with per-action metering models. Who is under pressure: Seat-based vendors without a meaningful consumption metric. Buyer impact: Software costs become variable and increasingly tied to AI utilization rather than headcount.
3. The Rise of the AI Agent
The 2026 enterprise software stack is being repriced in real time. At Knowledge 2026 in Las Vegas, ServiceNow unveiled Action Fabric, an integration layer that external AI agents must pass through to read data or execute workflows inside the platform. The action-based pricing model means customers pay per operation. The unit is the action, and the price scales with how extensively the agents work. This is not a one-vendor story. Workday's CEO publicly said agent metering offers significant upside for the company. HubSpot is reportedly moving in the same direction. SAP has taken a stricter posture by blocking unauthorized agents outright, while Datadog has capped agent traffic on its MCP server. The mechanisms differ, but the underlying recognition is the same: a SaaS pricing model built around human seats cannot fully capture the work AI agents perform.
GenAI may become the new user interface for enterprise software, according to Constellation Research. Enterprise software has historically been dominated by category leaders such as Salesforce in CRM, Workday in HR, Oracle and SAP in ERP, Microsoft in productivity, and ServiceNow in workflows and ITSM. In this model, GenAI becomes the new interface, while the underlying enterprise systems become the data and execution layer. ServiceNow, SAP, and Workday are consequently drawing new boundaries around customer data as external AI agents erode the traditional per-seat model.
4. From SaaS Applications to AI-Native Platforms
The so-called SaaSpocalypse framing highlights the scale of the potential disruption. Taskade's analysis identifies three pricing phases emerging as AI agents reshape enterprise software. In the first, per-seat pricing comes under pressure because AI agents can reduce the number of human users required. In the second, usage-based pricing remains resilient because AI agents increase consumption. In the third, ephemeral app generation allows AI to create bespoke applications on demand, potentially undermining the traditional economics of per-seat SaaS and affecting companies such as Salesforce, Workday, and Monday.com. Autonomous SRE swarms can also diagnose and remediate issues without relying on traditional dashboards, potentially affecting Datadog, PagerDuty, and Dynatrace. Meanwhile, zero-ticket enterprise AI could resolve IT and HR requests instantly, collapsing ticket queues and affecting platforms such as ServiceNow, Zendesk, and Freshworks.
For ServiceNow and Workday, the tollgate strategy turns disruption into a new revenue line. SAP has taken a stricter posture toward access attempts, seeking to route agent traffic through its own stack. Enterprise AI is replacing predictable per-seat billing with consumption models that behave less like subscriptions and more like utility invoices, leaving finance teams to manage spending that fluctuates with model activity rather than headcount.
5. Major Vendor Moves
- ServiceNow: ServiceNow introduced Action Fabric at Knowledge 2026, creating a layer through which external agents can execute governed workflows on the platform. The company's COO confirmed that activity will be metered, with customers paying per action. Its transition toward outcome-based pricing is establishing a new benchmark as concerns about seat-count erosion give way to a model in which providers are paid for the productivity AI agents deliver.
- SAP: SAP continues to use AI units as an underlying pricing currency. The company has also taken a stricter posture by blocking unauthorized agents and attempting to route agent traffic through its own stack. Joule sits on the permitted side of this model. SAP API Policy v4, Section 2.2.2, prohibits external AI agents from independently scheduling API calls without going through the SAP stack. The broader strategy reflects the emergence of enterprise software toll booths: a single agent can generate 50,000 API calls in a month, potentially turning previously invisible activity into a significant bill.
- Workday: Workday's CEO has said that charging for agent access offers considerable financial upside. The strategy mirrors ServiceNow's approach. Workday API behavior affecting roughly 30 enterprise companies, including Goldman Sachs, JPMorgan, and Salesforce, points to tighter governance enforcement. In HR and payroll, protection of personally identifiable information represents an additional potential moat.
- Salesforce: Salesforce is pursuing consumption-based pricing for Agentforce through several mechanisms, including approximately $2 per conversation, Flex Credits at about $500 per 100,000 credits, standard actions at roughly $0.10, and per-user add-ons ranging from $125 to $550 per month. Agentforce 1 Edition, priced at $550 per user per month, was introduced alongside these pricing options. Salesforce is therefore selling Agentforce through multiple pricing structures rather than a single model. More than 50% of Agentforce and Data 360 bookings reportedly came from existing customers, suggesting that buyers should plan for additive spending rather than assume AI will immediately replace existing software licenses. Salesforce has also raised prices on Slack and introduced additional Agentforce add-ons, reinforcing the shift toward consumption-based AI monetization.
- Oracle & Microsoft: Traditional vendors such as Microsoft, SAP, Oracle, and IBM continue to leverage large installed bases of on-premises customers while pushing toward SaaS-based consumption models. Cloud-native vendors including Workday, Zendesk, and ServiceNow continue to expand enterprise SaaS spending. The broader enterprise AI opportunity is increasingly being monetized through agent usage rather than traditional per-seat fees.
6. CRM, ERP & Business Applications
CRM and ERP platforms are increasingly becoming data stores surrounded by toll booths. GenAI is emerging as the new user interface, and without that interface layer, enterprise systems increasingly function as underlying data and workflow infrastructure. Salesforce Agentforce, particularly its Service Agent offering, uses consumption-based pricing, including approximately $2 per conversation, Flex Credits at around $500 per 100,000 credits, standard actions at roughly $0.10, and per-user add-ons ranging from $125 to $550 per month. Agentforce Help Agent and Customer Service Portal are also moving toward pay-per-resolution pricing, tying the cost of the software more directly to customer outcomes.
In ERP, SAP uses AI units as an underlying currency, while its Joule assistant remains within the company's controlled access model. Oracle Integration provides native connectivity across Oracle and non-Oracle SaaS and on-premises environments, including Oracle ERP Cloud, Service Cloud, HCM Cloud, Salesforce, Workday, EBS, SAP, NetSuite, and ServiceNow. The integration layer is therefore becoming more than a connectivity layer. It is increasingly becoming an enforcement layer through which vendors can govern data access, workflow execution, and AI-agent activity.
7. Agentic Workflow & Automation
Zero-ticket enterprise AI could resolve IT and HR requests instantly, collapsing traditional ticket queues and putting pressure on platforms such as ServiceNow, Zendesk, and Freshworks. Ephemeral app generators could allow AI to create bespoke applications on demand, potentially undermining per-seat SaaS economics at companies such as Salesforce, Workday, and Monday.com. Autonomous SRE swarms could diagnose and remediate infrastructure issues without relying on traditional dashboards, potentially affecting Datadog, PagerDuty, and Dynatrace.
ServiceNow is betting heavily on enterprise AI with a vision of managing increasingly large portions of business operations. The company's approach illustrates the emerging revenue model for enterprise software: monetize the work performed by agents rather than simply the number of employees using the software. The broader market is moving in the same direction. ServiceNow's Action Fabric enables external agents to execute governed workflows, while Workday sees agent metering as a potentially significant source of upside. Datadog, meanwhile, has capped third-party agents at 5,000 daily requests against its MCP server. The mechanisms differ, but the underlying strategy is increasingly clear: control how agents access enterprise systems, then monetize the activity they generate.
8. Data, Integration & Governance
The emergence of agent-based pricing means enterprise customers may soon be charged for individual data queries and actions across SaaS platforms. The unit is the action, and the price scales with how extensively agents work. This is not a one-vendor development. Workday sees agent metering as a potential source of upside, HubSpot is reportedly moving in the same direction, SAP has taken a stricter posture by blocking unauthorized agents, and Datadog has capped agent traffic on its MCP server. The mechanisms differ, but the underlying recognition is consistent: a SaaS pricing model built around human seats cannot fully account for the activity generated by autonomous agents.
Security and governance are becoming central to this transition. Salesforce focuses on Agentforce CRM data isolation, Workday on HR and payroll PII protection, Microsoft on Copilot and Azure AI enterprise integration security, Oracle on Fusion AI and database access to financial systems, and ServiceNow on Now Assist security across ITSM workflows. The broader governance gap remains significant: developers can build side-project agents, departments can connect agents to external tools, and SaaS vendors can embed agentic features, while many organizations still lack clear inventories and governance processes even as agents gain access to sensitive enterprise data.
9. SaaS Economics & AI Pricing
The reported $285 billion market rout following Anthropic's agent releases in early 2026 was not simply panic; it reflected a market learning how to price the potential impact of ephemeral app generators, autonomous SRE swarms, and zero-ticket enterprise AI. Per-seat pricing, based on a fixed monthly fee for each human user, faces existential pressure if AI agents reduce seat counts. Usage-based models appear more resilient because AI agents increase consumption.
Salesforce has pursued multiple AI pricing structures. Agentforce can be purchased through consumption-based mechanisms including approximately $2 per conversation, Flex Credits at about $500 per 100,000 credits, standard actions at roughly $0.10, and per-user add-ons ranging from $125 to $550 per month. SAP continues to use AI units, while ServiceNow combines seat-based pricing with secondary usage metrics. The common direction is clear: AI activity is becoming a measurable economic unit rather than an invisible extension of a traditional software license.
More than 50% of Agentforce and Data 360 bookings reportedly came from existing Salesforce customers. That suggests customers are not necessarily replacing Salesforce licenses with Agentforce but purchasing AI capabilities in addition to existing software. Buyers should therefore plan for additive spending rather than assume immediate substitution savings. Agentforce 1 Edition is priced at $550 per user per month, reinforcing the need for CIOs and CFOs to model both seat-based and consumption-based AI expenses.
10. M&A & Competitive Landscape
The reported $500 billion repricing of enterprise SaaS illustrates how quickly AI expectations are changing the competitive landscape. Salesforce, Workday, and Oracle fell together as investors reassessed the assumption that AI would simply enhance existing software. The alternative scenario is more disruptive: AI agents could replace portions of the work performed through enterprise applications, forcing vendors to defend their economics while simultaneously creating new monetization models.
Anthropic's partnership with Blackstone to build a new services company for integrated AI products and services further illustrates the broader commercialization of enterprise AI. Meanwhile, ServiceNow's Action Fabric provides a practical example of how enterprise software vendors can respond to the agentic shift: control the access layer, govern the actions, and charge for the work performed through the platform.
11. What It Means for CIOs
- Spending: Enterprise AI is replacing predictable per-seat billing with consumption models that behave more like utility invoices. Finance teams will need to manage spending that fluctuates with model activity rather than headcount. The emerging pricing unit is the action, while Agentforce and similar platforms introduce additional consumption metrics. With more than 50% of Agentforce bookings reportedly coming from existing customers, CIOs should budget for additive AI spending rather than assume immediate license substitution.
- Buying behavior: A single AI agent can generate 50,000 API calls in a month. Under traditional per-seat pricing, that activity might have generated no incremental charge; under consumption models, it could generate thousands of dollars in costs. Datadog's 5,000-request daily cap for third-party agents demonstrates why CIOs need clarity around AI units, Flex Credits, per-action pricing, per-conversation charges, and pay-per-resolution models before committing to large-scale deployments.
- Governance: Vendors are drawing new boundaries around customer data as external AI agents gain access to enterprise systems. SAP has taken a stricter posture by blocking unauthorized agents and routing permitted activity through its own stack. ServiceNow governs workflows through Action Fabric, while Workday sees agent metering as a potential revenue opportunity. CIOs need agent inventories, permission mappings, and governance policies before connecting autonomous systems to critical business applications.
- Architecture: GenAI is becoming the new interface for enterprise applications, while systems such as Salesforce, Workday, Oracle, SAP, Microsoft, and ServiceNow increasingly function as the underlying data and execution layer. The result is a new architecture in which vendors can build toll booths around the access, execution, and data layers used by AI agents.
12. The CODEW Analysis
Enterprise software is undergoing its third major rewrite. The first moved from client-server architectures to the cloud between 1999 and 2010. The second moved from cloud software toward mobile and cloud-native architectures between 2010 and 2022. The third, now underway, is the transition from SaaS applications to an agentic workforce between 2024 and 2030. Each rewrite changes the pricing model, competitive moat, and balance of power between vendors and buyers. The second era rewarded distribution, allowing companies such as Salesforce, ServiceNow, and Workday to build powerful businesses around software seats. The third era is likely to reward execution: the ability to turn intelligence into measurable work, deliver ROI, manage permissioned data, and execute auditable actions—and then charge for that activity.
The August 27 signal is therefore the emergence of the toll booth rather than the seat. ServiceNow, SAP, and Workday are all developing mechanisms that allow them to control or monetize AI-agent activity. ServiceNow's Action Fabric lets external agents execute governed workflows, with customers paying per action. Workday sees considerable upside in charging for agent access. SAP is taking a stricter approach by attempting to route agent activity through its own stack. Datadog has capped third-party agent traffic on its MCP server. The mechanisms differ, but the underlying recognition is the same: a SaaS pricing model built around human seats cannot fully capture the work performed by AI agents.
The reported $285 billion market rout following Anthropic's agent releases in early 2026 was therefore less about panic than about the market learning how to price a new software economy. Per-seat pricing faces pressure if agents reduce human seat counts, while usage-based pricing becomes more resilient as agents increase consumption. The vendors best positioned for the next cycle will likely be those that solve agent identity, map human permissions to agent permissions, enforce least privilege, provide observability and cost governance, and monetize activity through mechanisms such as pay-per-resolution, pay-per-action, AI units, or Flex Credits. Vendors that simply sell agents as expensive add-ons without solving governance and economic predictability may face greater cancellation risk. The interface is no longer the application. It is the agent—and the agent now pays the toll.
13. What to Watch Next
- Action Fabric enforcement: Does ServiceNow's per-action metering become an industry standard, or will the market remain divided between SAP's blocking approach and Workday's more permissive agent-access model?
- AI units vs. Flex Credits vs. per-action pricing: Will SAP's AI units, Salesforce's Flex Credits, and ServiceNow's per-action model converge, or will fragmented pricing structures create a new layer of FinOps complexity for enterprise buyers?
- The 50,000-API-call problem: Will the economics of a single agent generating 50,000 API calls per month force CIOs to implement agent inventories, permission mapping, and governance controls before connecting autonomous systems to critical business applications?
- The $500 billion repricing: Will the reported market-value losses across Salesforce, Workday, and Oracle become a permanent valuation reset from the thesis of software enhanced by AI toward software potentially replaced by AI agents?
- Pay-per-resolution: Will Salesforce's pay-per-resolution approach prove that outcome-based pricing can scale across enterprise software, or will buyers continue to prefer seat-based AI licensing with secondary consumption metrics?
Reviewed by Erwin Castro
on
Thursday, August 27, 2026
Rating: