Agentic AI is not just the next buzzword—it represents a transformative shift in enterprise technology. Unlike static AI models that respond only to specific prompts, agentic AI operates with a degree of autonomy. These agents can reason, plan, and execute tasks on behalf of users, offering dynamic interactions across a company’s digital environment.
As business complexity increases and digital transformation continues to accelerate, agentic AI promises to offload cognitive labour, enhance productivity, and streamline decision-making processes. This promise is reflected in major platform announcements, particularly from SAP and Microsoft, who have integrated agentic capabilities into their enterprise suites.
Agentic AI refers to systems designed to act as autonomous agents that can understand context, set goals, make decisions, and execute actions independently. These systems go beyond task-specific AI to simulate human-like agency and collaboration.
Unlike traditional AI, which requires explicit commands, agentic AI systems can proactively identify problems, formulate plans based on organizational priorities, coordinate with humans or other agents, and learn from user behaviour and outcomes. This level of capability changes how software is used. Instead of static interfaces, users can delegate entire workflows to their AI counterparts.
SAP introduced Joule in 2023, positioning it as a generative AI copilot embedded across the SAP ecosystem. Joule’s objective is to simplify user interactions with ERP systems, surface insights from complex datasets, and proactively assist in workflow automation.
In 2024 and 2025, SAP has updated Joule with increasingly agentic features. Joule now acts as a workflow broker across SAP S/4HANA, SuccessFactors, Ariba, and SAP Analytics Cloud. It adapts its recommendations based on user roles, historical data, and real-time analytics. In procurement scenarios, Joule can auto-approve low-risk purchases within predefined thresholds and flag anomalies for review. These enhancements align with SAP’s broader vision of embedding AI deeply into every business process.
SAP has shifted to a consumption-based licensing model for AI capabilities. Clients purchase “AI Units” which are deducted based on the intensity of Joule’s usage (e.g., queries, workflows, integrations). As AI usage scales, so does the cost—creating both flexibility and uncertainty for enterprises.
SAP’s official pricing guide (as of 2025) outlines AI Units akin to compute tokens, with overages billed in tiered rates. Importantly, clients must budget for AI capacity in their annual planning to avoid unanticipated spikes.
Microsoft 365 Copilot, first released commercially in 2023, has become a key AI layer across Office applications. Microsoft’s implementation is distinctly agentic in its ability to summarize meetings, draft complex documents, manipulate Excel formulas, and even automate email triage.
Recent updates from Microsoft Build 2025 highlight new multi-agent collaboration features where Copilot can persist memory across sessions, assign sub-tasks to other embedded agents, and integrate with Microsoft Teams for cross-user task coordination.
Microsoft’s licensing model remains premium-priced: Copilot is available as a $30 to $42.50/month add-on to M365 E3/E5 or Business Standard plans. For an enterprise of 1,000 users, this could translate to an annual Copilot bill upwards of $500,000.
This raises key questions: Are AI enhancements delivering enough ROI to justify these costs? Can this model scale sustainably?
In finance, Joule can run automated reconciliations and flag inconsistencies in GL accounts. Copilot can generate financial statements and analyse YOY variances without manual input.
In HR, SAP SuccessFactors with Joule can recommend hiring actions based on workforce analytics. Copilot can draft employee letters, analyse survey data, and suggest training plans.
In supply chain management, Joule acts on predictive inventory models, automatically initiating replenishment workflows. Copilot analyses vendor performance data and suggests renegotiation terms.
The critical question for CIOs and procurement leaders is whether these tools provide enough quantifiable benefit to offset their substantial cost.
Potential value drivers include time savings by reducing report generation or documentation by 80% in some cases. They also improve decision accuracy by leveraging real-time data for faster, better outcomes. Additionally, agentic AI empowers employees by enabling non-technical staff to perform complex tasks.
However, Gartner warns of “AI inflation,” where vendors add marginal AI functionality to justify pricing increases. CIOs must assess if these capabilities meaningfully reduce operating costs or improve service delivery.
A U.S.-based pharma firm piloting Copilot across 500 users reported a 13% increase in productivity in knowledge roles—but only 6% cost savings due to overlapping functions with existing systems.
The rise of agentic AI is complicating traditional software license negotiations. Historically, enterprise agreements were based on modules, users, and CPU cores. AI introduces variables such as API call volume, task completion thresholds, and data residency for AI training.
Agentic AI introduces risks, particularly in regulated industries. Financial services firms must ensure that AI-driven decisions (e.g., credit recommendations) comply with transparency and auditability mandates. Joule and Copilot must be configured to log decisions, provide justification paths, and operate within ethical guardrails.
In the EU, the upcoming AI Act will likely require companies to conduct risk assessments and register high-risk AI systems. Enterprises using Joule or Copilot in sensitive workflows will need compliance mappings and impact assessments.
Tom Davenport, Distinguished Professor at Babson College, states: “Agentic AI represents a paradigm shift. It’s no longer about tool augmentation—it’s about outcome delegation.”
Satya Nadella, CEO of Microsoft, observes: “We will soon expect AI agents to do 60-70% of the planning and administration behind tasks we barely think about today.”
Fei-Fei Li, Professor of Computer Science at Stanford University, emphasizes: “Trust and transparency will define the next decade of enterprise AI. Without them, even the best tools will go unused.”
Procurement and IT leaders navigating this shift should begin by evaluating the maturity and autonomy of the AI systems under consideration. Does the agentic AI operate autonomously or require frequent prompts? Is the model auditable and explainable? How does licensing scale with enterprise-wide usage?
Contractually, buyers should include clauses for AI performance SLAs, negotiate test-and-adapt periods with usage data reviews, and require transparency into future AI model changes.
From a governance perspective, organizations should establish an AI risk committee, create feedback loops between business units and IT, and track AI tool usage versus productivity metrics quarterly.
To make informed decisions, organizations must understand how licensing scales with user count. For instance, a mid-sized company with 200 users purchasing Microsoft Copilot at $42.50/user/month would face an annual cost of $102,000. If the same company uses SAP’s Joule and consumes a moderate volume of AI Units, it might incur an additional $30,000 to $50,000 annually based on estimated usage patterns, bringing total AI tool expenditure to around $150,000 per year.
In contrast, a large enterprise with 10,000 users adopting Copilot at the same rate would see an annual cost of $5.1 million. SAP Joule’s usage at this scale, especially across diverse operations such as HR, finance, and logistics, could lead to a bill between $1.5 million to $3 million depending on AI Unit consumption. Altogether, AI-driven licensing could exceed $8 million per year for large organizations.
These figures underscore the importance of understanding both fixed licensing rates and variable consumption costs when evaluating enterprise AI tools.
Agentic AI marks the beginning of a profound evolution in enterprise software. SAP’s Joule and Microsoft’s Copilot show how AI agents can augment human capacity, simplify interactions, and optimize operations. But as enterprises evaluate these tools, key questions remain: Is licensing models mature and predictable enough for enterprise scale? Will productivity gains remain consistent across diverse teams? Can governance and risk management keep pace with AI’s rapid evolution?
For now, enterprises must proceed with both ambition and caution. Those that master the agentic AI lifecycles election, licensing, deployment, and governance—will define the next generation of operational excellence.
2Data is an independent software licensing advisory specializing in Microsoft, Oracle, SAP, Salesforce and IBM optimization. We help companies reduce contract costs, manage risk, and simplify their software licensing.
