By 2026, every major ERP vendor has announced their “agentic AI” solution. SAP has Joule, Microsoft has Copilot for Dynamics 365, Oracle has its Fusion AI Agents, Sage deploys Sage Copilot, Odoo released its AI App, and UK-based Access Group is preparing its AI assistant. The marketing message is unified: the agent that understands, reasons, and executes. The product reality varies from a few months to 18 months gap between the pitch and general availability.
This article cuts through the noise, vendor by vendor, to reveal what’s actually deployable in production in Q1 2026, what remains in early adopter mode, and what’s still just Gartner slideware. The dual objective: enable a CIO or CFO to assess whether embedded AI should factor into their ERP choice today, and give business leaders a concrete view of what will change in their daily operations. We’ve selected six major vendors (SAP, Microsoft, Oracle/NetSuite, Sage, Odoo) plus three challengers (Access Group UK, IFS, Unit4), with a synthesis table and explicit recommendation by company profile.
Definition: What is “agentic AI” in ERP really
The term agentic AI is not synonymous with generative AI. A classic LLM responds, summarizes, drafts. An agent executes multi-step actions in a target system, with an objective, a plan, and often a human validation workflow. Applied to an ERP, this ranges from simple (“create this quote from this email”) to sensitive (“validate and release this manufacturing order if stock and planning conditions are met”).
We can classify ERP copilots in 2026 into three maturity levels:
- Suggestion: the assistant proposes text, a label, a line to complete. Systematic human validation. This is the level of most features marketed as “AI” in 2024.
- Guided execution: the agent pre-fills a complete action (bank reconciliation, accounting entry, purchase order) that the user validates with one click.
- Autonomous execution: the agent triggers the action if business conditions are met, with audit and human escalation at threshold. This is the level that Oracle Fusion AI Agents and SAP Joule Studio aim for in 2026, on bounded domains.
The distinction from ERP chatbots of three years ago is real: those responded to FAQ in natural language. A 2026 agent has read/write access to business objects and can orchestrate a sequence. This changes the CIO’s central question: it’s no longer “does the agent understand my question”, but “who is attributed the risk when it makes a mistake”. An agent that wrongly closes an accounting period, releases a manufacturing order on incorrect stock, or sends a collection reminder to a strategic prospect is not a simple UX defect. It’s a business responsibility, often legal, that must be traced, audited, and assumed.
SAP Joule
Status Q1 2026. Joule is integrated into SAP S/4HANA Cloud Public Edition since the November 2024 release (SAP News, December 2024). Joule Studio, which allows IT teams to build their own agents, went GA, and SAP announces for Q1 2026 the general availability of agents like the Bid Analysis Agent and Production Planning and Operations Agent, the latter capable of validating and releasing manufacturing orders if stock/planning conditions are met (aimultiple, SAP AI Agents 2026).
What it actually does. Data query responses (“What cost variances on Q1 by site”), entry generation, assistance with purchase and sales order creation, supplier document synthesis. For finance, Joule feeds variance tables and proposes reasoned commentary.
LLM model. SAP uses multi-model orchestration via Generative AI Hub (OpenAI, Anthropic, Mistral, AWS Bedrock depending on task) rather than an in-house LLM.
Pricing and packaging. Native access for S/4HANA Cloud Public Edition customers. AI consumption is billed in AI Units, with public pricing around €7 per unit with an annual floor, to be confirmed in contract. On-premise customers are not served: SAP confirmed Joule is not on the classic systems roadmap.
Limitations. Dependency on S/4HANA Cloud deployment, no access for large ECC bases still in production. Quality heavily depends on master data cleanliness. In English, finance/controlling use cases are better covered than supply chain.
For whom. Mid-market and large accounts already on S/4HANA Cloud Public Edition. If you’re still on ECC, Joule is not a lever for 2026.
Microsoft Copilot for Dynamics 365
Status Q1 2026. The most mature on the market in customer execution. Copilots for Sales and Service have been GA since 2024, and the 2026 release wave 1 of Dynamics 365 emphasizes agentic innovations across sales, service, finance, supply chain, HR and commerce (Microsoft Dynamics Blog, release wave 1 2026).
Strong use cases. Copilot for Sales writes opportunity summaries, prepares meetings, generates follow-ups. Copilot for Service solves level 1 tickets by tapping into the knowledge base. Copilot for Finance (in preview) automates reconciliation and variance analysis. Integration with Microsoft 365 (Outlook, Teams, Excel) is the real differentiator.
Pricing. Microsoft 365 Copilot for Sales and Copilot for Service are billed $50/user/month standalone, or $20/user/month for customers who already have a Microsoft 365 Copilot subscription (Microsoft 365 Copilot pricing). Some baseline Copilot capabilities are included in Dynamics 365 Sales SKUs (Dynamics 365 Sales pricing).
Limitations. Value collapses outside Microsoft ecosystem: if your users have neither Outlook nor Teams, half of Copilot scenarios lose their interest. Data governance is strong on the Microsoft 365 side but requires minimum E3/E5 licensing for advanced functions.
For whom. Companies already invested in Microsoft ecosystem. For a CIO standardizing on Azure and Microsoft 365, Copilot D365 becomes almost a no-brainer. For an SME under Google Workspace, the ROI is much less obvious.
Oracle Fusion AI Agents and NetSuite
Status Q1 2026. Oracle has been communicating since 2024 about a massive agent library. In October 2025, the company claimed over 600 embedded intelligent agents (400 in Oracle Fusion Cloud Applications, 200 in Oracle Industry Applications), with Fusion Agentic Applications officially announced in March 2026 (Oracle, Fusion Agentic Applications, March 2026).
Concrete examples. Financial Close Agent that drives period closure subledger by subledger, Expense Management Agent that controls and codes expense reports, Procurement Agent that recommends suppliers. On the NetSuite side, integration is more progressive but the roadmap aligns.
LLM model. Oracle primarily uses Cohere and in-house models for certain sensitive uses (finance), with hosting on Oracle Cloud Infrastructure.
Pricing. Fusion agents are included in Oracle Cloud Applications SKU at their GA release, without surcharge (Oracle AI Agents for Fusion Applications). The AI Agent Marketplace adds a layer of partner agents (Accenture, Deloitte, IBM, Stripe, Box) with specific pricing.
Limitations. Functional depth varies greatly from one agent to another. Multilingual tooling remains uneven, English is covered but with delay compared to US releases. For NetSuite, many agents are still in ramp-up.
For whom. Large accounts and North American mid-market already on Oracle Fusion or NetSuite. The zero surcharge promise is a strong negotiation argument against SAP and Microsoft.
Sage Copilot
Status Q1 2026. Sage Intacct 2026 Release 1, delivered February 13, 2026, embeds a wave of AI functions, including Sage Copilot and Finance Intelligence Agent in early adopter (Rand Group, Sage Intacct 2026 Release 1). Sage Copilot is presented as a generative assistant dedicated to finance (Sage Copilot, Sage US).
Strong use cases. Automatic vendor invoice extraction (vendor, amount, date, lines), Close Assistant that tracks closure status subledger by subledger in real time, Variance Analysis that flags budget/actual variances with generative explanation. Bank reconciliation largely automated.
Product scope. Sage Intacct priority (top of portfolio), progressive deployment on Sage 200 and Sage 50 cloud, integration in Sage for Accountants (UK). International markets follow the roadmap with a few months delay.
Pricing. Some AI functions are included by default, others require configuration or early adopter program. Final packaging of Finance Intelligence Agent is not public at this stage, to be requested under NDA.
Limitations. Explicit target: finance. Few embedded supply or commerce scenarios by Sage Copilot. The agent remains early adopter on its most agentic capabilities: production customers rely mainly on level 1 (suggestion) at end Q1 2026.
For whom. Accounting firms and finance teams of SME-mid-market under Sage Intacct. For small business under Sage 50, impact in 2026 remains limited to extraction and reconciliation level.
Odoo AI (Odoo 18 and 19)
Status Q1 2026. Odoo introduced its first AI functions in version 18 (lead scoring, invoice OCR, email generation), then accelerated with Odoo 19 delivered April 2025, which includes a dedicated AI App, customizable AI agents, natural language search in database, and meeting transcription (Certum Solutions, Odoo 19 AI Features). The AI App allows a non-developer administrator to build an agent.
Real use cases. Email and ticket summary, lead scoring, bank reconciliation, invoice extraction, automatic response based on knowledge base. Agentic logic (chaining actions in Odoo) is present via the AI App but remains to be built by the customer, Odoo does not deliver finished agents equivalent to SAP or Oracle.
Pricing. Native AI functions (Lead Scoring, AI Fields, OCR, AI App) are reserved for Odoo Enterprise licenses. Token quotas to be verified in contract. Possibility to self-host a local model for data-sensitive companies.
Limitations. Real gap with SAP/Microsoft copilots on agentic maturity. Odoo bets on self-service: you build your agents, unlike Oracle which delivers 400 out-of-the-box. The advantage is flexibility, the drawback is the effort required.
For whom. SMEs and scale-ups under Odoo Enterprise that have internally minimum technical skills to configure the AI App. Gains on repetitive tasks (entry, classification, scoring) are real. Don’t count on Odoo AI for an accounting closure agent equivalent to Sage Close Assistant.
The challengers: Access Group AI, IFS.ai, Unit4 ERPx
Access Group AI. The UK-based ERP vendor (Access Dimensions, Access Financials, Access Workspace) launched its AI initiative in 2024 with focus on mid-market businesses. AI capabilities include document processing, automated approval workflows, and predictive analytics. Early 2026 deployment on Access Dimensions with strong emphasis on UK regulatory compliance.
IFS.ai. IFS pushes asset-intensive oriented AI (manufacturing, FSM, energy). Less visible in the 2026 generalist debate but relevant in its verticals.
Unit4 ERPx. The Dutch vendor targets service organizations with AI for project accounting, resource planning, and financial consolidation. AI assistant in beta across multiple products in Q1 2026.
These players are not at SAP or Microsoft level on autonomous agentic AI, but they matter if your scope is UK/EU focused, mid-market, or sector-specific.
Global comparison table
| Vendor | Agentic AI Maturity (Q1 2026) | Strong Use Cases | Indicative Pricing | Underlying LLMs | License Prerequisites |
|---|---|---|---|---|---|
| SAP Joule | Level 2 to 3, Joule Studio GA, finance/supply agents in GA | Finance, supply chain, purchasing | AI Units (~€7/unit), partially included S/4HANA Cloud | Multi-model via GenAI Hub | S/4HANA Cloud Public Edition |
| Microsoft Copilot D365 | Level 2, very mature on sales/service | Sales, Service, Finance (preview) | $50/user/month ($20 with M365 Copilot) | OpenAI via Azure | Dynamics 365 + Microsoft 365 |
| Oracle Fusion AI Agents | Level 2 to 3, 600+ embedded agents | Finance (closure), expense, procurement, HR | Included in Oracle Cloud Applications | Cohere + in-house models | Oracle Fusion / NetSuite |
| Sage Copilot | Level 1 to 2, Finance Intelligence Agent in early adopter | Finance, closure, invoices | Included or add-on by module | OpenAI (via Azure) | Sage Intacct priority |
| Odoo AI | Level 1 to 2, agents to build via AI App | Scoring, OCR, summaries, custom agents | Included Odoo Enterprise, quotas to verify | OpenAI (self-host option) | Odoo Enterprise |
| Access Group AI | Level 1, gradual deployment 2025-2026 | Document processing, approvals, compliance | To request | Not public | Access Dimensions, Financials |
The 4 questions to ask your vendor before signing
1. Does the copilot work on production data or on a frozen sandbox? Some agents are trained and tested on demo data sets. In reality, they stumble on business cases that your instance has and the demo doesn’t. Request a POC on your scope.
2. Where does the data flow and where is it hosted? EU or US? Azure, AWS, OCI? Which LLM, what retention, what reversibility? This is the central question for GDPR and sovereignty. Check server address and applicable law, not just the vendor name.
3. What is the granularity of roles and permissions on the agent? An agent that validates an accounting entry must be governed at the same level as a human user. If the vendor responds “the agent has the rights of the user who launches it”, ask about the technical account behind and audit traces.
4. What quota, what monthly limit, what behavior at saturation? A year-end closure can multiply by ten the volume of calls to the agent. If your contract is capped at 100 AI Units per year, you have a problem. Negotiate thresholds and overrun effects.
Recommendations by profile
Industrial mid-market already under SAP S/4HANA Cloud. Joule is a credible lever from 2026 for finance and supply. If you’re still on ECC, agentic AI won’t be an argument before your S/4HANA migration.
Large account under Oracle Fusion or NetSuite. The zero surcharge promise on 400+ native agents is strong. Treat it as a negotiation advantage to leverage at each renewal.
SME or mid-market already invested in Microsoft 365. Copilot for D365 combined with Microsoft 365 Copilot offers the best value/effort ratio thanks to Outlook, Teams, Excel integration. The $20/user/month for M365 Copilot customers makes entry more accessible.
Accounting SME under Sage Intacct. Sage Copilot and Finance Intelligence Agent will genuinely change the daily life of closure and AP teams from this year, with a clearly assumed finance focus.
Scale-up or SME under Odoo Enterprise. Odoo AI covers repetitive tasks well (OCR, scoring, summaries). To build a more advanced workflow agent, the AI App requires internal investment but opens the door without changing ERP.
UK SME attached to a national vendor. Access Group AI is worth watching closely, still in gradual deployment. If you’re signing an Access Group contract in 2026, demand written commitment on the 2026-2027 AI calendar rather than a commercial pitch.
Conclusion
Agentic AI embedded in ERP is not in 2026 a decisive criterion for most mid-market choices: maturity is uneven, pricing opaque, independent ROI absent. But the gap between vendors is widening, and the decision made today on an ERP commits your ability to leverage it in 2027-2028, when agents will reach level 3 (autonomous execution) on bounded processes. Anticipating means demanding today the dated roadmap, license conditions, and data sovereignty clause. It’s also refusing the marketing figure “+50% productivity” as long as no independent study supports it on an ERP scope. The subject being in rapid movement, the editorial position to favor holds in three words: skepticism, POC, contracts. Skepticism toward keynote slides. POC on your scope with your data. Clear contracts on quota, data, reversibility.
To go further, consult our top 5 AI solutions for ERP in 2026 which covers complementary third-party AI building blocks, our Odoo vs SAP vs NetSuite comparison for the vendor ecosystem, and our dedicated analysis Odoo vs Microsoft Dynamics 365 to go deeper on two of the vendors compared here.
