Your finance team is spending hours every week on work that software should be doing. Invoice intake. Bank reconciliation. Sales order processing from emails. Inventory monitoring. These are not high-value tasks. They are necessary, repetitive, and expensive in staff time.
Microsoft has been building toward a different model for Business Central. Not just automation, which ERP systems have always had in some form. Something more capable: AI agents that can read documents, interpret context, take action inside the ERP, and route exceptions to humans, all without requiring a fixed rule set for every scenario.
As of May 2026, that shift is well underway. The 2026 Release Wave 1, rolling out from April to September 2026, frames Business Central explicitly as an “agentic ERP platform.” That is not marketing language. The feature changes behind it are substantive, and decision-makers need an honest understanding of what has actually shipped, what is in preview, and what the real operational and governance implications are.
This guide covers all of it.
What Is Microsoft Dynamics 365 Business Central?
Before going deep on AI agents, a baseline is useful for readers evaluating this platform for the first time.
Microsoft Dynamics 365 Business Central is a cloud-based ERP system built for growing and mid-market organizations. It consolidates finance, supply chain, inventory, sales, purchasing, and project operations into a single platform. Because it sits within the Dynamics 365 family, it connects natively with Microsoft 365, Power BI, Power Platform, Dataverse, and other Dynamics 365 applications.
Business Central is available primarily as a SaaS product (Business Central online) and is where all AI and Copilot development is focused. On-premise and private cloud versions exist but receive limited or no AI agent functionality.
The platform serves organizations from around 10 employees to several hundred, and is particularly well-suited for businesses that are already embedded in the Microsoft ecosystem.
Understanding the Difference: Copilot vs. AI Agents
This distinction matters more in 2026 than it did a year ago, because Microsoft now ships both within Business Central, and they serve genuinely different purposes.
Copilot is the AI-powered assistant layer. It operates when a user initiates an interaction. You type a question, ask for a summary, request line suggestions on a sales order, or ask Copilot to help reconcile a bank account. Copilot responds. The human is always in the driver’s seat. Copilot is included in standard Business Central licenses for most tenants.
AI agents operate differently. They are autonomous workers that monitor data, detect events, interpret documents, and take action without waiting for a user to prompt them. They run in the background, process transactions around the clock, and route exceptions to humans only when needed. Agents are designed for high-volume, repetitive processes where continuous monitoring has more value than on-demand assistance.
The practical distinction looks like this. Copilot helps a finance user reconcile a bank account by suggesting matches. An AI agent monitors the accounts payable inbox overnight, extracts invoice data from PDFs, creates draft purchase invoices, and presents them for human approval the next morning, without anyone asking it to.
Both are valuable. They are not interchangeable.
The table below maps the core differences:
| Dimension | Copilot | AI Agents |
|---|---|---|
| Activation | User-initiated | Event-driven, time-based, or data-triggered |
| Interaction model | Conversational | Background processing with approval routing |
| Automation depth | Advisory and assistive | Semi-autonomous to autonomous within limits |
| Human involvement | Every interaction | Exception handling and approval only |
| Ideal for | Productivity, navigation, content generation | High-volume, repetitive operational processes |
| Setup required | Minimal, available by default | Requires configuration and governance design |
What the 2026 Release Wave 1 Changes
Microsoft’s 2026 Release Wave 1 for Business Central, running April through September 2026, is the most significant AI update the platform has received. The stated direction is explicit: the wave accelerates “the move to agentic ERP.”
Key changes in this wave include:
Immersive Home workspace. A new AI-powered adaptive landing page that surfaces actionable recommendations, agent activity, and workflow monitoring in a single view. It functions as an agent management and operational intelligence hub, making it easier for users to see what agents are doing and where human input is needed.
Expanded Payables Agent capabilities. The Payables Agent now supports invoice-to-purchase-order matching before goods receipt and can trigger receipt posting automatically when conditions are met. This reduces the manual touchpoints in the accounts payable cycle further than the original version allowed.
Agents supporting bookkeeping tasks. Agents can now assist with bookkeeping tasks based on recognized patterns from historical transaction behavior. This is an expansion of the financial agent scope from pure document processing toward pattern-informed categorization.
Model Context Protocol (MCP) server improvements. The wave includes updates to MCP server integration, enabling AI agents built in Copilot Studio to connect with Business Central data more reliably and with better cross-application context. This is the foundation for more sophisticated custom agent builds.
Power Platform integration depth. Tighter Dataverse and Power Automate integration gives AI agents the ability to span Business Central and connected systems with less custom development effort.
Custom agent development toolkit (preview). Microsoft has released an AI Development Toolkit in preview that allows developers and consultants to design, test, and iterate custom agents in sandbox environments. This is covered in detail later in this guide.
These are not minor feature additions. They represent a genuine architectural shift in how Business Central handles operational workflow automation.
Core AI Agent Capabilities: What Is Live Today
Payables Agent
The Payables Agent is the most mature autonomous AI agent in Business Central as of May 2026. It handles the accounts payable intake process end to end, with human approval as the final gate before posting.
The agent works by monitoring a designated mailbox for incoming vendor invoices. When a PDF invoice arrives, the agent imports it, runs AI-based document analysis to extract the relevant fields (vendor, invoice number, line items, amounts, VAT), and creates a draft purchase invoice in Business Central. It applies historical matching data to propose the correct posting accounts for unmatched lines, and routes the draft to a human reviewer before anything is posted.
Key capabilities in the current version:
- Continuous mailbox monitoring, no manual import step required
- AI-based extraction from invoice PDFs, including non-standard formats
- Automatic matching to existing purchase orders and receipts where applicable
- Invoice-to-purchase-order matching before goods receipt (new in 2026 Wave 1)
- Auto-triggered receipt posting when matching conditions are met (2026 Wave 1)
- Multi-attachment handling from a single email, processed as separate invoices
- Exception routing for invoices the agent cannot match with sufficient confidence
- Full audit trail of agent actions for review and compliance purposes
For organizations processing more than 50 vendor invoices per month, this agent produces a measurable reduction in AP processing time. The agent handles the intake and matching work. The human handles approvals and exceptions.
Sales Order Agent
The Sales Order Agent automates the front end of the order fulfillment process. It monitors a shared inbox for customer requests, interprets what the customer is asking for, and builds a sales quote using live Business Central data.
The full processing sequence the agent handles:
- Reads incoming customer emails requesting quotes or orders
- Identifies the customer by matching sender details to Business Central records
- Interprets product requests, including vague or informal descriptions
- Checks inventory availability and validates delivery dates
- Builds a complete sales quote with pricing, taxes, and delivery terms
- Drafts a professional email response with the quote as an attached PDF
- Routes the draft to a human reviewer for approval before sending
- Converts approved quotes to sales orders and prepares order confirmations
The critical design choice here is the approval gate. The agent does not send quotes autonomously. It prepares everything and presents it for human review. This matters both for accuracy and for compliance. A sales quote is a commercial commitment. The agent handles the time-consuming preparation. The human makes the commitment.
For businesses receiving multiple customer inquiries per day via email, the Sales Order Agent eliminates the manual work of interpreting the email, checking availability, building the quote document, and drafting the reply. That is typically 15 to 30 minutes of work per inquiry. At volume, the time savings are significant.
Copilot for Finance: The Assistive Layer
Distinct from the autonomous agents, Copilot provides a set of AI-assisted features that activate when a user requests them. These are live in current Business Central and included in standard licenses.
Bank account reconciliation assist. Copilot proposes matches between bank statement transactions and ledger entries and suggests posting accounts for unmatched lines. The human reviews and approves. This compresses the reconciliation process significantly, particularly for businesses with high transaction volumes.
E-document matching. For incoming vendor e-invoices, Copilot maps invoice lines to purchase order lines, proposes matches, and flags exceptions. This handles the matching complexity that would otherwise require manual line-by-line review.
Chat with Copilot. Users can ask natural language questions against Business Central data. “What is our current outstanding balance with Vendor X?” or “Which customer invoices are more than 60 days overdue?” gets answered directly, without building a report. This is one of the highest-adoption features because it removes the dependency on IT for routine data queries.
Analysis assist. Natural language commands can transform a list of records into an analysis view with grouping, pivots, and filters, without leaving Business Central or building a report in Power BI.
Summarize. Copilot provides role-relevant summaries of open records, surfacing what is urgent and important so users can act without reviewing every field.
Autofill. On document and card pages, Copilot suggests field values based on company data, historical patterns, and in some cases web lookups, reducing manual data entry.
Copilot for Inventory and Supply Chain
Demand forecasting. Business Central includes a non-generative AI forecasting model that predicts future demand based on historical sales data. This is used to drive replenishment suggestions, helping organizations avoid both stockouts and overstock situations. The forecasting model draws on transaction history and seasonal patterns, making it more responsive than static reorder points.
Item substitute suggestions. When items are out of stock or discontinued, Copilot analyzes product data and proposes compatible alternatives inline, so sales and service teams can keep orders moving without escalating to a purchasing or inventory manager.
Sustainability journal emissions. For organizations tracking greenhouse gas emissions, Copilot estimates total emissions based on input descriptions and adjusts emission factors for future entries.
Copilot for Sales and Marketing
Sales line suggestions. Write a few words describing what a customer wants, paste in a conversation, or upload a document, and Copilot identifies the matching items and quantities to add to a sales quote or order. This reduces the time spent building sales documents from customer communications.
Marketing text suggestions. Copilot drafts product descriptions based on item attributes already in Business Central and a defined tone preference. For businesses using the Shopify integration, this copy can be published directly to the online store.
Copilot for Company Setup
Number series suggestions. For new implementations or expansions to new entities, Copilot proposes consistent numbering formats for documents and master records based on existing conventions, reducing the risk of number range collisions.
The AI Development Toolkit: Building Custom Agents (Preview, May 2026)
One of the most significant developments in the 2026 wave for organizations with specific operational requirements is the AI Development Toolkit, currently in preview.
This toolkit allows product owners, consultants, and developers to design, configure, test, and iterate custom AI agents inside Business Central sandbox environments. The agents are built using natural language instructions rather than code, which lowers the barrier for non-developers to prototype agent behavior. Completed agents can then be graduated to production-ready AL extensions by developers.
The lifecycle Microsoft has defined for custom agents follows this sequence:
- Create. Define the agent identity, purpose, and assigned profile.
- Configure. Write natural language instructions that define the agent’s behavior, scope, and boundaries. Grant the agent the minimum permissions needed for its function.
- Activate. Enable the agent in the sandbox and add tasks to test its behavior.
- Iterate. Test outputs, refine instructions, and adjust permissions.
- Graduate. Export the learnings and build a production version as an AL extension.
- Clean up. Deactivate and remove the prototype agent from the sandbox.
Important constraints to understand before planning around this capability:
- The AI Development Toolkit is currently preview only and available exclusively in sandbox environments, not in production.
- It currently supports English only. Use in other languages may produce inconsistent results.
- Agents cannot use the Tell Me search function. Navigation is limited to what their assigned profile allows.
- Agents can only interact with one list on any given page. Pages with multiple sublists present a known limitation.
- Agents cannot have more permissions than the user who created them.
For organizations that want to prototype before committing to custom development, this toolkit is a meaningful capability. For those expecting to deploy custom agents to production immediately, the preview status and sandbox-only restriction are important constraints to factor into planning timelines.
Custom agent development also connects to Copilot Studio, where agents can be designed with conversational interfaces and connected to Business Central data through Power Platform and the MCP server integration introduced in the 2026 wave.
How AI Agents Work Technically: What Decision-Makers Need to Know
Business Central’s AI agents are powered by Azure OpenAI Service. Processing occurs in Microsoft’s cloud infrastructure. Data from the Business Central environment is sent to Azure OpenAI for inference, and results are returned to the application.
Three practical implications follow from this architecture:
Data residency. Regulated industries or organizations in regions with strict data sovereignty requirements must verify that their Business Central tenant configuration and Azure region assignment comply with applicable requirements. Microsoft publishes geographic availability and data residency boundaries for Copilot features. Verify current status before activating agents in environments handling sensitive data.
SaaS dependency. AI agents and Copilot capabilities are designed for Business Central online. On-premise deployments have limited or no access to these features. Organizations on on-premise versions who want AI agent functionality need to evaluate the migration path to Business Central online before planning agent deployments.
Data quality dependency. Agents reason from ERP data. Bank reconciliation agents match against ledger entries. Sales Order Agents pull inventory and pricing from live records. Payables Agents apply historical matching patterns. All of this depends on the quality and completeness of the underlying data. Poor master data produces unreliable agent outputs. This is not a theoretical concern. Organizations with recent implementations, inconsistent data entry practices, or incomplete vendor and customer records will find agent accuracy lower than expected until data quality is addressed.
The permissions model is also worth understanding. Agents operate with explicitly assigned permission sets. Following the principle of least privilege, an agent should have access only to the data and functions it actually needs. This is both a security best practice and a practical accuracy measure. An agent with overly broad permissions sees more contextual noise, which can reduce output quality.
Pricing: What AI Agents Actually Cost
Business Central’s AI agents use a consumption-based pricing model tied to Microsoft Copilot Credits. Each Copilot Credit costs $0.01. Credits are consumed when agents perform AI-driven actions, such as interpreting a document, checking a record, or generating a response.
Standard Copilot features available to all Business Central online users, including Chat, Analysis Assist, Summarize, and bank reconciliation assist, are included with Business Central licenses at no additional consumption cost.
AI agents that perform autonomous processing consume Copilot Credits based on the complexity and volume of their actions.
Payables Agent cost estimate:
A typical invoice processed by the Payables Agent involves approximately 65 Copilot Credits, based on document interpretation, line matching, and draft creation steps. At $0.01 per credit, processing 100 invoices per month costs approximately $65. Processing 500 invoices per month costs approximately $325.
Emails without valid invoice attachments do not consume credits.
Sales Order Agent cost estimate:
A typical order processing sequence, including email interpretation, inventory check, quote creation, and response drafting, consumes approximately 16 to 17 Copilot Credits per request. At 100 customer requests per month, the estimated cost is approximately $16.50. At 500 requests, approximately $82.50.
Requests involving attachments with usable order data consume additional credits.
Custom agents built in Copilot Studio follow the same Copilot Credits model. Cost is driven by the complexity and frequency of the agent’s actions.
Several important caveats apply:
- These estimates are based on Microsoft’s published credit consumption figures and are subject to change.
- Advanced Copilot capabilities tied to Microsoft 365 Copilot licensing are separate from Business Central Copilot and carry additional per-user costs.
- Organizations should validate current licensing and pricing directly with Microsoft or a qualified partner, as this area continues to evolve.
- Consumption billing for agents requires explicit setup through the tenant admin center. Billing is not activated by default on all tenants.
The cost structure is, by any measure, low relative to the staff time these agents replace. The more relevant cost consideration for most organizations is implementation, governance design, and change management, not the per-credit consumption fee.
Use Cases by Business Size
Small Business (Under 50 Employees)
For small businesses, the most constrained resource is always staff time. Every hour a finance team member spends on invoice intake or bank reconciliation is an hour not spent on analysis, customer relationships, or strategic work.
The highest-value AI agent applications at this size:
- Payables Agent to handle vendor invoice intake without manual processing.
- Bank reconciliation Copilot to reduce month-end close time.
- Chat with Copilot to give business owners direct access to operational data without reports.
- Sales Order Agent for businesses that receive customer inquiries by email.
The business case is straightforward at this scale. The implementation complexity is low, the consumption costs are modest, and the productivity return is direct and measurable.
Mid-Market Business (50 to 500 Employees)
Transaction volume at this size makes manual processing costly in aggregate. AI agents address the cost of scale, not just individual task efficiency.
Priority applications:
- Payables Agent across higher invoice volumes where AP headcount would otherwise need to grow.
- Sales Order Agent for inside sales teams handling significant email inquiry volume.
- Demand forecasting to manage inventory across more complex product ranges.
- Analysis Assist and Chat with Copilot to reduce report dependency across departments.
- Custom agents for industry-specific or role-specific processes where Microsoft’s standard agents do not cover the scenario.
The 2026 wave’s Immersive Home workspace becomes particularly relevant at this size, providing a consolidated view of agent activity and exception queues that operations managers and finance directors need to maintain oversight as automation scope expands.
Enterprise or Multi-Entity Organizations
Business Central is not typically the primary ERP for large enterprises, but it is frequently deployed for specific business units, subsidiaries, or regional operations. In these configurations, AI agents address high-volume processing within the unit while connecting to broader enterprise workflows through Power Platform and Dataverse.
Key applications at this scale:
- Intercompany transaction matching and reconciliation across multiple Business Central entities.
- Consolidated demand forecasting across inventory locations.
- Automated variance analysis in management accounts.
- Agent-to-agent coordination using the multi-agent capabilities now generally available in Copilot Studio.
Governance is the dominant consideration at this scale. Clear audit trails, approval thresholds, escalation paths, and compliance documentation are prerequisites before expanding agent autonomy. The 2026 wave’s governance enhancements in Power Platform and Copilot Studio admin controls address part of this requirement, but organizational governance design remains the primary work.
Industry-Specific Applications
Manufacturing
Manufacturers benefit most from AI agents in production-adjacent scenarios. Demand-driven replenishment agents that connect inbound order signals to material requirements and vendor lead times reduce both planning cycle time and the cost of material shortages or excess stock. The 2026 wave’s enhanced purchasing automation, including pre-receipt invoice matching, is directly relevant to manufacturing operations with complex goods receipt workflows.
Distribution and Wholesale
For distributors, inventory intelligence is the most direct value driver. Agents that continuously monitor inventory aging, flag slow-moving SKUs before they become write-offs, and adjust replenishment suggestions based on actual sell-through data rather than static thresholds address one of the core margin risks in distribution. The Sales Order Agent is also high-value for distributors whose customers send purchase orders by email, a common scenario in B2B distribution.
Professional Services
Services businesses using Business Central for project accounting benefit from AI agents that monitor project margin in real time, surface budget variance early, and assist with time and expense matching. The Chat with Copilot feature reduces the dependency on custom reporting for project financial oversight, which is particularly useful for project managers who need quick access to financial data without waiting for a finance report run.
Retail and eCommerce
Retailers benefit from demand forecasting agents and, for those using the Shopify integration, from the marketing text generation Copilot feature that can populate product descriptions at scale. Inventory replenishment agents that respond to actual sell-through data rather than seasonal assumptions are directly relevant for retailers managing trend-sensitive product ranges.
Key Decision Factors Before Activating AI Agents
These are the questions that determine whether an AI agent deployment delivers value or creates noise.
Is the underlying data clean? AI agents reason from ERP data. If vendor master records are incomplete, if invoice matching has historically been inconsistent, or if chart of accounts coding is irregular, agents will produce unreliable outputs. A data quality audit is a prerequisite, not a parallel workstream.
Are the relevant processes defined? Agents work best in documented, repeatable processes. An informal purchasing process where different team members follow different steps does not become consistent by adding an agent. It produces inconsistent agent behavior. Process standardization comes before agent deployment.
Is the governance model ready? Who reviews agent outputs? Who approves exceptions? What happens when an agent produces an incorrect suggestion? Who owns the configuration? These questions need organizational answers before the first agent goes live, not after the first error.
What is the user adoption plan? Finance and operations staff who are accustomed to manual workflows do not automatically trust AI-generated suggestions. Without clear training on what the agent does, why its outputs should be trusted within defined parameters, and how to handle exceptions, adoption stalls. Change management investment is consistently the most underestimated cost in AI agent deployments.
What are the compliance requirements? In regulated industries or for organizations subject to audit requirements, AI-generated transactions and automated postings require additional documentation and approval steps. Finance teams should review compliance implications before activating any automated transaction features, particularly the auto-posting capabilities introduced in the 2026 wave.
Is the tenant on Business Central online? If the organization is still on an on-premise Business Central version, AI agent features are not available. The decision to activate agents and the decision about ERP deployment model are linked.
Red Flags and Common Deployment Mistakes
These are patterns that consistently undermine AI agent deployments in Business Central implementations.
Activating agents without an exception review process. Agents route exceptions for human review. If there is no defined process for handling those exceptions, the queue builds up and gets ignored. Unreviewed exception queues mean the agent is producing work nobody is acting on.
Treating preview features as production-ready. The AI Development Toolkit is in preview as of May 2026. Preview features in Business Central are explicitly not intended for production use, are subject to change, and carry supplemental terms of use. Organizations that deploy preview features in production environments take on support and stability risk that Microsoft does not cover under standard service agreements.
Expecting value without data preparation. Demand forecasting agents need at least two to three years of clean sales history to produce reliable outputs. Payables Agents produce better matching results with well-maintained vendor master data. Deploying agents on thin or dirty data produces suggestions that mislead rather than assist.
Ignoring regional feature availability. Not all Copilot and AI agent features in Business Central are available in all regions at the same time. Microsoft releases features on a rolling basis. Organizations in the Middle East, Africa, and parts of Asia Pacific should verify current availability before planning around specific capabilities.
Underestimating agent permission design. An agent configured with broad permissions will interact with data and UI elements it does not need. This creates noise in agent reasoning and increases the risk of unintended actions. Agents should be configured with the minimum permissions required for their defined function, using purpose-built profiles rather than general user profiles.
Not monitoring agent performance after go-live. AI agents should be reviewed periodically for output quality, exception rates, and processing accuracy. An agent that was accurate at go-live may drift in accuracy if data patterns change or if process changes are made without updating agent instructions.
What AI Agents in Business Central Do Not Do
This matters as much as the capability list.
AI agents do not replace human judgment on complex decisions. They handle processing and pattern recognition. Strategic decisions, vendor negotiations, pricing strategy, and capital allocation remain human responsibilities.
They do not guarantee accuracy. An AI agent can misread an invoice, propose an incorrect GL account, or misidentify a customer. Human review checkpoints exist for this reason. Removing those checkpoints to accelerate throughput transfers the risk of agent errors into posted transactions.
They do not work well in disorganized data environments. The quality of agent outputs is bounded by the quality of the data they work with.
They are not a shortcut to ERP maturity. If the Business Central implementation is unstable, if workflows are not defined, or if user adoption of the base system is low, adding AI agents adds complexity without adding value. The foundation must be solid first.
They are not universally available. Feature availability varies by region, license type, and deployment model. Confirming what is available in a specific tenant before building plans around it is essential.
The 2026 Broader Context: Microsoft’s Agent 365 and What It Means for Business Central Users
In May 2026, Microsoft announced Microsoft Agent 365, a centralized governance layer for managing AI agents across the Microsoft 365 and Dynamics 365 ecosystem. The announcement repositioned AI agents as the foundational operating layer for Microsoft’s productivity stack, not just individual features within specific applications.
For Business Central users, this has practical implications. The Microsoft 365 2026 Work Trend Index found that nearly half of Copilot chat activity now supports what Microsoft classifies as high-value cognitive work, including analysis, decision-making, and problem-solving. The company’s direction is clear: AI agents are being positioned as the mechanism through which organizations scale output without scaling headcount.
Copilot Studio’s multi-agent coordination capabilities, now generally available as of early 2026, allow agents built across different Microsoft applications to coordinate with each other using Agent-to-Agent (A2A) protocols. This means a Business Central AI agent handling invoice processing can communicate with an agent in Outlook handling vendor communications, creating coordinated multi-step workflows that span application boundaries.
For Business Central decision-makers, this broader context means two things. First, the platform investment in AI agents is long-term and increasing. Choosing Business Central now means choosing a platform that will continue to receive significant AI capability investment. Second, the governance and organizational design work required to use these capabilities responsibly is expanding alongside the capability set. Organizations that build sound governance frameworks now will be better positioned to expand agent scope as the platform matures.
Conclusion
AI agents in Business Central have moved from proof of concept to operational reality. The 2026 Release Wave 1 deepens the Payables Agent, introduces expanded purchasing automation, adds the Immersive Home workspace for agent oversight, and opens a developer toolkit for custom agent design. These are not incremental improvements. They represent a genuine architectural shift in what a mid-market ERP can do.
The organizations that benefit most from these capabilities share a consistent profile. They are on Business Central online, they have invested in data quality, their core processes are documented, and they have a governance model for reviewing and approving agent outputs. Without those foundations, AI agents generate noise rather than signal.
The path forward is methodical. Audit current data quality. Identify the two or three processes where high volume, high repetition, and time sensitivity make agent support most valuable. Design the governance model before activating automation. Start with the built-in agents, measure their impact, and expand from there. Custom agent development through Copilot Studio or the AI Development Toolkit makes sense once the standard agents are performing reliably and the organizational readiness to manage more complex automation is established.
Businesses that approach AI agents as a capability to be built incrementally on solid foundations will extract lasting value. Those that treat activation as the endpoint will be disappointed by the results.
What are AI agents in Microsoft Dynamics 365 Business Central?
AI agents in Business Central are autonomous software workers that monitor data, interpret documents, and execute or propose actions inside the ERP without requiring user prompts for every step. They are powered by Azure OpenAI Service and operate through Microsoft’s Copilot framework. Current production agents include the Payables Agent for accounts payable automation and the Sales Order Agent for email-based order processing. A developer toolkit for custom agent design is available in preview as of May 2026. Agents are available on Business Central online only.
What is the difference between Copilot and AI agents in Business Central?
Copilot in Business Central is an AI assistant that responds when a user initiates a request, helping with tasks like bank reconciliation, natural language queries, sales line suggestions, and data analysis. AI agents operate autonomously in the background, triggered by events or schedules, processing invoices, handling order intake, and monitoring data without waiting for user prompts. Copilot is included in standard Business Central licenses. Agents consume Copilot Credits at $0.01 each, based on usage volume.
What does the Business Central 2026 Release Wave 1 add for AI agents?
The 2026 Release Wave 1, running April to September 2026, expands the Payables Agent to support invoice-to-purchase-order matching before goods receipt and enables auto-triggered receipt posting. It introduces the Immersive Home workspace for agent activity monitoring, adds bookkeeping pattern-based agent assistance, and improves MCP server integration for custom agent development. The wave also includes a preview AI Development Toolkit for building and testing custom agents in sandbox environments using natural language instructions.
How much do AI agents cost in Business Central?
Business Central AI agents use a consumption-based model. Each Copilot Credit costs $0.01. The Payables Agent consumes approximately 65 credits per invoice, making the cost around $65 for 100 invoices per month. The Sales Order Agent consumes approximately 16 to 17 credits per customer request, costing around $16.50 for 100 requests per month. Standard Copilot features such as Chat, Analysis Assist, and bank reconciliation are included in Business Central licenses at no additional consumption cost. Licensing structures are evolving and should be verified with Microsoft directly.
Can I build custom AI agents in Business Central?
Yes. Custom AI agents can be built through two paths. The AI Development Toolkit, currently in preview, allows design and testing of custom agents in sandbox environments using natural language instructions, without requiring coding. Agents can be graduated to production as AL extensions. Copilot Studio provides a second path for building conversational and task-based agents connected to Business Central data through Power Platform and the MCP server integration. Both approaches follow the Copilot Credits consumption pricing model.
FAQ Section
Are AI agent features available in all countries?
No. Microsoft releases Business Central Copilot and AI agent features on a rolling regional basis. Availability in the Middle East, Africa, and parts of Asia Pacific may lag behind North American and European markets. Current regional availability is published in Microsoft’s Copilot International Availability report and should be checked before planning deployments around specific features.
Does using AI agents in Business Central require additional licensing beyond the standard subscription?
Standard Copilot features are included in Business Central Essentials and Premium licenses for eligible tenants. AI agents that perform autonomous processing consume Copilot Credits at $0.01 each, billed based on usage. Consumption billing requires explicit setup through the tenant admin center. Advanced capabilities connected to Microsoft 365 Copilot or Agent 365 may require separate licensing. Current terms should be verified with Microsoft or a qualified partner.
Can I use the AI Development Toolkit in a production Business Central environment?
No. The AI Development Toolkit is currently a preview feature and is available exclusively in sandbox environments. Microsoft explicitly states that preview features are not intended for production use and may have restricted functionality subject to change. Organizations should complete agent design and testing in sandbox environments before building production-ready versions as AL extensions.
How does Business Central protect data when AI agents process documents?
Business Central’s AI features process data through Microsoft’s Azure OpenAI Service under Microsoft’s enterprise data protection commitments. Customer data is not used to train shared AI models. Data residency is governed by the Azure region configuration of the Business Central tenant. Organizations in regulated industries or with data sovereignty requirements should review Microsoft’s data processing documentation and confirm regional configuration before activating AI agents.
What happens when an AI agent makes an error?
Business Central’s agent design includes human review checkpoints before consequential actions like posting or sending documents externally. The Payables Agent creates draft invoices for human approval before posting. The Sales Order Agent presents quotes for review before sending. Exceptions that fall outside the agent’s confidence threshold are routed to human review queues. No agent in Business Central posts transactions fully autonomously without a defined approval path. Organizations should establish clear processes for handling exception queues as part of deployment planning.
Is the Sales Order Agent suitable for businesses where customers send orders in many different formats?
The Sales Order Agent uses AI-based document and email interpretation, so it handles variation in customer language and request format better than rules-based automation would. However, it performs best when customer data is well-maintained in Business Central and when item descriptions in emails have reasonable overlap with item catalog data. Highly informal or non-standard requests will produce more exceptions requiring human review. The agent is suited for organizations where email order intake is a significant volume activity and where most customers follow reasonably consistent request formats.
