Agentic Arbitrage: When AI Agents Replace Classic Enterprise Software

What Is Agentic Arbitrage—and Why Does Gartner Call It the 'Saaspocalypse'?

Agentic arbitrage describes the mechanism by which AI agents autonomously complete tasks across multiple systems—without users needing to open traditional interfaces. Gartner VP George Brocklehurst puts it succinctly: 'Agentic AI changes the economics of software. 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.'

Concretely: instead of an employee logging into SAP S/4HANA to create a purchase order, switching to Salesforce to check customer status, and then requesting approval in Microsoft Teams, an AI agent orchestrates these steps in the background. The agent accesses all three systems via API, executes the transaction, and logs it—without human interaction with the interfaces. Result: the number of required named-user licenses falls, concurrent licenses become obsolete, and the ROI of the original software investment shifts.

The Numbers: Agentic AI Growing at Triple-Digit Rates

Global AI spending will reach $2.59 trillion in 2026 according to Gartner (+47 percent year-on-year). Within this volume, AI agent software records the steepest growth:

• 2026: $206.5 billion (+139 percent from $86.4 billion in 2025)
• 2027: forecast $376.3 billion (+82 percent)
• 2029: $752.7 billion at a compound annual growth rate (CAGR) of 119 percent

For context: only 17 percent of organizations have deployed AI agents in production to date; over 60 percent plan a rollout within the next two years. By end of 2026, 40 percent of all enterprise applications will integrate task-specific agents—up from under 5 percent in 2025 (Gartner, August 2025).

Impact on SAP, Salesforce, and Microsoft Investments

SAP and Claude Integration

At Sapphire 2026, SAP announced that Anthropic's Claude will be the primary reasoning and agentic capability of the SAP Business AI Platform. Claude agents will orchestrate tasks in SAP S/4HANA, SuccessFactors, and Ariba via Model Context Protocol (MCP)—within the same governance controls that apply to human decisions. Over 100,000 organizations already use Claude models on AWS Bedrock.

For Swiss SAP customers, this means: existing named-user licenses may be partially replaced by agent-based access. Instead of licensing per head, billing may shift to API calls, transaction volume, or outcomes. Contracts negotiated two years ago do not reflect these usage scenarios.

Salesforce and Microsoft: Similar Dynamics

Salesforce is also integrating agents (Einstein Copilot), and Microsoft is expanding Copilot across Azure, Dynamics 365, and Microsoft 365. The logic remains the same: agents navigate across these systems and execute workflows that previously required multiple human logins. Licensing models based on seat counts come under pressure; simultaneously, the need for API capacity, governance tooling, and monitoring rises.

The Governance Gap: Adoption Outpaces Control by Factor of 8

While adoption of agentic AI explodes, governance lags massively behind. Gartner warns: over 40 percent of agentic AI projects will be canceled by end of 2027 due to escalating costs, unclear business value, or inadequate risk controls. Further findings:

• Only 7 to 8 percent of enterprises have mature agent governance (FifthRow/Enterprise Agentic AI Landscape 2026).
• Only 23 percent can fully inventory and trace agent actions (identity/governance gap).
• Agentic AI adoption outpaces governance maturity 8 to 1 (Gartner Cybersecurity Report).
• Enterprises spend 17 times more on AI tools than on securing AI itself (Gartner 2026).
• Production agent rollouts are being pulled back at high rates; PII exposure and hallucination cited as leading failure causes (Gartner/TELUS Digital/Sinch, June 2026).

Swiss Compliance: revDSG and EU AI Act

For Swiss companies, the situation is compounded by regulatory requirements. The revised Data Protection Act (revDSG, in force since September 2023) mandates that automated decisions be disclosed and subject to human review. AI agents that autonomously trigger orders, adjust contracts, or change credit limits fall under this provision.

Organizations with EU market exposure must also consider the EU AI Act. High-risk AI systems—for example in HR or credit decisioning—are subject to strict documentation, transparency, and audit obligations from August 2, 2026. Agentic AI operating in such domains must be classified and controlled accordingly.

Budget Recalibration: From Seat Model to Outcome Model

The central question for CIOs and CFOs is: how much of today's SaaS spending can be substituted by agentic AI—and what new cost blocks arise?

New Cost Structures

1. Agent software and platform licenses: API-based pricing models, often billed by tokens, transactions, or outcomes.
2. Inference costs: Claude Sonnet 5 costs, for example, $2 per million input tokens (promotional price through end of August 2026); production deployments with high throughput generate substantial ongoing costs.
3. Governance and monitoring: tools for agent observability, audit trails, PII scanning, hallucination detection.
4. Integration and orchestration: middleware to securely connect agents with legacy systems (e.g., SAP, Salesforce, ERP).
5. Change management and training: employees must learn to work with and alongside agents.

Evaluate Savings Realistically

Not every license can be eliminated immediately. Many applications are deeply embedded in business processes; compliance, audit, and reporting functions often require human approvals. Moreover, experience shows that initial agent deployments are iterative—see 'Von der Pilot-Falle zum ROI: Wie Schweizer KMU KI-Agenten erfolgreich skalieren.' Plan with an 18 to 36 month horizon to realize meaningful license reductions.

Action Recommendations for Swiss Decision-Makers

1. Audit software portfolio: identify applications with high license burden and repetitive, rule-based tasks—ideal candidates for agent substitution.
2. Renegotiate contracts: demand API-friendly pricing models and flexibility for seat reductions. Anchor clauses for bot/agent usage.
3. Establish governance: implement agent IAM, audit logging, and escalation mechanisms before you scale. Verify revDSG and EU AI Act compliance.
4. Pilot and measure: start with clearly scoped use cases, define success criteria, and measure ROI quantitatively—our article 'KI-ROI messen: Der CFO-Leitfaden für Schweizer Unternehmen' offers metrics and frameworks.
5. Evaluate vendors: SAP/Claude, Microsoft Copilot, Salesforce Einstein—compare platforms on governance maturity, compliance support, and total cost of ownership.
6. Change management: communicate transparently which tasks agents will handle and how roles will evolve. Agentic AI is not purely an IT topic but concerns the entire organization.

Agentic AI changes the economics of software. Agentic systems deliver outcomes directly, bypassing traditional UX-heavy applications. This breaks the link between user growth and revenue growth.

Conclusion: Arbitrage as Strategic Opportunity—Not Threat

Agentic arbitrage is neither hype nor distant future. With a market volume of $206.5 billion in 2026 alone and a CAGR of 119 percent through 2029, the trend is set. For Swiss enterprises, this means: existing enterprise software investments must be scrutinized, governance gaps closed, and budget allocations recalibrated. Those who act early can optimize licenses, increase efficiency, and ensure compliance. Those who wait risk competitors realizing arbitrage advantages first—or governance failures leading to costly rollbacks.

KI-Outsourcing.ch supports Swiss organizations in strategically planning agentic AI, implementing it in compliance with governance requirements, and integrating it into existing enterprise landscapes—as an external AI unit that combines operational excellence with regulatory rigor.