In 2024 the buzzword in enterprise software was “copilot”, an assistant that answers questions. In 2026 it is “agent”: AI that takes actions inside business systems, such as creating a sales order, matching an invoice or resolving a ticket. SAP, Oracle, Microsoft and ServiceNow all shipped agents this year, and the main differences lie in data strategy, pricing and governance. For consultants, the skills that matter now are process redesign, agent configuration, data quality and AI governance.
This is analysis based on vendor announcements and independent reporting up to September 2026. Vendor performance figures are the vendors’ own claims.
What happened in 2026: a timeline
| Date | Vendor | Announcement |
|---|---|---|
| 26 Feb | ServiceNow | Unveils “Autonomous Workforce”: role-based AI specialists, starting with a Level 1 IT service desk specialist |
| 24 Mar | Oracle | Announces Fusion Agentic Applications: 22 new agentic apps planned with release 26B, plus consumption-based pricing |
| 27 Mar | SAP | Discloses acquisition of Reltio (master data management) |
| 4 May | SAP | Announces deals for Dremio (data lakehouse) and Prior Labs (tabular AI models) |
| 5 May | ServiceNow | At Knowledge 2026, extends AI specialists to IT, CRM, employee services and security |
| May | SAP | At Sapphire 2026, launches the Autonomous Suite: 50+ Joule assistants orchestrating 200+ agents |
| 25 Aug | Microsoft | Ends Dynamics 365 and Power Platform release waves; roadmap moves to the continuous AI at Work roadmap |
Vendor by vendor
SAP: Joule assistants on top of a data platform
At Sapphire in May, SAP presented more than 50 domain-specific Joule assistants for finance, supply chain, procurement, HR and customer engagement, which coordinate a pool of over 200 agents. Joule Studio 2.0 lets developers build agents in Python or with AI coding tools, and an AI Agent Hub, announced for the third quarter of 2026, is meant to govern SAP and non-SAP agents in one place. SAP said RISE with SAP customers get three Joule assistants activated in year one, while GROW with SAP customers get the full agent portfolio.
The bigger story is data. SAP’s three 2026 acquisitions all feed its Business Data Cloud: Reltio for master data, Dremio for querying data in open Apache Iceberg tables without copying it into SAP, and Berlin-based Prior Labs for foundation models built for tables of business data. Financial terms were not disclosed. SAP’s chief executive, Christian Klein, has said 80% accuracy is not good enough for payroll, the financial close or supply chain planning, which explains the focus on clean, connected data.
For SAP teams still on ECC, these features are another reason the migration question cannot wait; see our SAP ECC maintenance deadline explainer.
Oracle: agents inside Fusion transactions
Oracle’s Fusion Agentic Applications embed agents in transactional workflows with the aim of running whole processes, such as reducing days sales outstanding. AI Agent Studio adds a natural-language builder and an ROI dashboard that tracks time and cost saved per agent. Customers can set guardrails and require human approval.
On pricing, Oracle is moving from per-user subscriptions to consumption-based “action units” of roughly one cent each, with an initial allotment included. Earlier generative AI features and agents remain at no extra cost. This is usage-based pricing, not payment by business outcome, even though Oracle markets the agents around outcomes.
Microsoft Dynamics 365: Copilot agents and a continuous roadmap
Microsoft’s agents act on everyday business documents. In Business Central, for example, the Sales Order Agent reads customer emails in a Microsoft 365 mailbox and drafts quotes and sales orders, with configurable human review. It consumes Copilot Credits, so admins set daily email limits and watch usage. Custom agents are built in Copilot Studio.
In August, Microsoft retired the twice-yearly “release wave” model for Dynamics 365, Power Platform and Dataverse. New capabilities now appear on the AI at Work roadmap as they are committed, and the transition is due to finish by 15 November 2026. Microsoft says products with established release schedules keep them, and Message Center remains the source for tenant-specific changes.
ServiceNow: AI specialists for defined job roles
ServiceNow’s Autonomous Workforce gives each AI specialist a defined role and the same permissions a human in that role would hold. The first, a Level 1 service desk specialist, handles tasks such as password resets and access requests and passes harder cases to human L2 or L3 staff. ServiceNow says it handles over 90% of its own employee requests internally. By May, the specialists covered IT, CRM, HR, legal, finance and procurement services, with security specialists due for general availability in September. AI Control Tower, which governs agents, is included across ServiceNow products. Our ServiceNow Autonomous Workforce guide covers the career angle in detail.
Side-by-side comparison
| SAP | Oracle | Microsoft | ServiceNow | |
|---|---|---|---|---|
| Agent model | Joule assistants orchestrating agents | Agentic apps inside Fusion | Copilot agents in Dynamics 365 | Role-based AI specialists |
| Build tool | Joule Studio 2.0, BTP | AI Agent Studio | Copilot Studio | AI Agent Studio |
| Pricing signal | Bundled with RISE/GROW offers | Action units (usage) | Copilot Credits (usage) | AI credits (usage) |
| Data bet | Business Data Cloud plus 2026 acquisitions | Fusion data and OCI | Dataverse and Fabric | Data already in its workflow platform |
| Governance | AI Agent Hub (Q3 2026) | Guardrails, human approval | Admin controls, credit limits | AI Control Tower |
Common themes, and the caveats analysts raise
- Data comes first. Agents act on master data, open orders and knowledge bases. Analysts covering Oracle and ServiceNow both warned that messy production data means humans will often need to step in.
- Supervised autonomy. HFS Research’s Ashish Chaturvedi expects “supervised autonomy” to be the norm for the next 24 months, not fully autonomous processes.
- Pricing is harder to forecast. Usage-based credits replace predictable seat licences. Info-Tech’s Scott Bickley warned that AI credits threaten budget predictability.
- Vendors grade their own agents. ROI dashboards and “hours saved” figures come from the vendors, so clients need independent measurement.
- Cloud editions get the features. The newest agent capabilities are tied to cloud subscriptions, which adds pressure on companies still running older on-premise ERP.
What consultants and developers should learn
| Role | What changes | What to learn next |
|---|---|---|
| Functional consultant | Designing who does what: agent, approver, exception handler | Process mapping, approval thresholds, testing agent output against business rules |
| Technical consultant | Building and connecting agents | Joule Studio and BTP, Copilot Studio, Oracle AI Agent Studio, ServiceNow AI Agent Studio; REST APIs and integration |
| Data and MDM | Agents expose bad master data quickly | Master data governance, data quality rules, lakehouse basics |
| Security and GRC | Agents are new identities with permissions | Least-privilege access, audit trails, prompt-injection risks |
| FinOps and licensing | Usage-based AI pricing | Tracking credits and action units, forecasting AI spend |
A practical way to start: pick one process you know well, such as order-to-cash, and write down every step, decision and exception. Then mark which steps an agent could do, which need human approval, and what data each step depends on. That exercise is close to what clients will pay for. The security side is covered in our guide to securing AI agents at work.
Will agents replace ERP jobs?
Routine, rules-based work such as L1 tickets, simple data entry and standard reports is most exposed; ServiceNow has said openly that its first specialist targets L1 service desk work. But each agent has to be designed, configured, tested, governed and maintained, and analysts expect humans to stay in the loop for exception-heavy and regulated processes. Consultants who combine domain process knowledge with agent configuration skills are best placed.
Related on this site
- Oracle’s $638 billion backlog and what it means for Oracle professionals
- ServiceNow’s Autonomous Workforce and AI Control Tower
- How to learn SAP in 2026
Sources
- ServiceNow: Autonomous Workforce expansion (5 May 2026)
- Microsoft: Dynamics 365 joins the AI at Work roadmap (25 August 2026)
- Microsoft Learn: Business Central Sales Order Agent
- CIO: SAP Sapphire 2026 agent announcements
- CIO: Oracle Fusion Agentic Applications and pricing
- TechTarget: SAP acquisitions of Reltio, Dremio and Prior Labs
Last reviewed: 8 October 2026.

Leave a Reply