Microsoft Dynamics 365 Business Central Edition
Insights, updates, and practical guides for Microsoft Dynamics 365 Business Central
AI in Business Central 2026: The Agents Changing Your ERP
- September 17, 2026
- 9:51 am
For twenty years, the operating assumption behind every Dynamics 365 Business Central deployment, and Navision before it, was that the system records what people do. A sales coordinator reads the customer’s email, interprets it, keys the order, checks stock, and releases the document. Business Central’s job was to hold the record and enforce the rules. The intelligence was always human; the software was the ledger.
That model persists because it works, because teams are trained on it, and because every workflow, approval matrix, and job description in a Business Central shop assumes a person is the actor. Its cost is equally familiar: order desks that spend their mornings re-keying emails, AP clerks matching invoices to receipts line by line, and controllers who know the close would be faster if someone, anyone, could pre-validate the routine 80% of transactions before a human ever looked at them.
That assumption formally broke with the 2025 release wave 2, and the 2026 release wave 1 (version 28, rolling out April through September 2026) is where agentic Business Central matures into production infrastructure. The Payables Agent is now generally available with 3-way matching, multi-language support, and Known Senders acceleration.
A new expense agent entered public preview in May 2026, capturing receipts by mobile and categorizing expenses with AI. Companies can now run multiple sales order agents per company with smarter item matching.
The MCP server gained major enhancements, telemetry, and a connection into the admin center itself. And Microsoft shipped the governance surface agents were missing: a dedicated cross-agent task pane, on-page review of agent-generated content, created-by-AI indicators, and an instant stop control for all active agent tasks.
Agents are not Copilot with a longer leash. Copilot responds when a user asks; an agent acts when a signal arrives (an email, a receipt, a transaction, or a schedule), reasons over the data within scoped permissions, executes multi-step work through the same UI a user would navigate, and escalates to a human only at the thresholds you define.
This article covers what AI in Business Central actually changes in 2026: the seven operational shifts the current agent lineup introduces, what agents cost to run, the prerequisites most organizations skip, and the difference between deploying agents deliberately and inheriting them by default at the next update.
Table Of Contents
- Quick Answer
- Defining AI in Business Central (Why the Distinction Matters)
- The 7 Shifts the 2026 Agent Lineup Introduces
- Copilot-Assisted Business Central vs. Agent-Operated Business Central
- Prerequisites Before You Activate Agents
- The Cost of Watching This Wave From the Sidelines
- How Omni Logic Delivers: The AI-First Framework
- Closing Perspective
- FAQs
Quick Answer
- Business Central 2026 release wave 1 (version 28, April to September 2026) makes agents production infrastructure: Payables Agent GA with 3-way matching, Expense Agent in public preview (May 2026), multiple Sales Order Agents per company (July 2026), and enhanced MCP server with telemetry and audit logs.
- Agents differ from Copilot architecturally: Copilot is interactive: request and response. Agents are autonomous: triggered by events, executing multi-step tasks in the background, with human-in-the-loop only where configured.
- The Sales Order Agent reads inbound customer emails, extracts line items, validates against live inventory, and drafts orders for review; 2026 wave 1 adds smarter item matching, relevance filtering, and multi-agent setups per company.
- The Payables Agent (GA) captures vendor invoices, runs 3-way matching against purchase orders and receipts, categorizes processed emails in Outlook, and accelerates trusted vendors through Known Senders.
- The Expense Agent (preview) captures receipts via mobile and web, categorizes and itemizes expenses with AI, and routes them through approval workflows.
- Governance shipped with the agents: a dedicated task pane for all agent tasks, on-page review of agent-generated content, created-by-AI record indicators, and an instant stop for all active agent tasks.
- Agents consume Copilot credits (prepaid or pay-as-you-go), so cost governance is a design input, not an afterthought.
- Omni Logic Solutions deploys these capabilities through its AI-First Framework: workflows classified as human, Copilot-assisted, or agent-automated before configuration begins, typically inside a ~90-day initial deployment.
Defining AI in Business Central (Why the Distinction Matters)
The phrase “AI in Business Central” now covers two architecturally different things, and conflating them is how deployments go wrong.
Copilot is AI-added interaction. It has been in the product for several release waves: chat over your data, drafting marketing text, bank reconciliation assistance, and autofill suggestions. It is valuable, but it is reactive. Nothing happens until a user asks, and the user remains the actor for every step. Most published commentary on Business Central AI agents still describes this layer and calls it agentic. It is not.
Agents are AI-first execution. An agent in the current architecture is a non-human actor inside the tenant: it has its own profile controlling which pages and actions it can see, its own permission sets scoped to least privilege, and its own audit trail. It interprets a goal, plans the steps, navigates Business Central pages the way a user would, and pauses for human approval only where its instructions require it. Microsoft’s own framing in the 2026 Wave 1 release plan is explicit: agents “handle more complex, end-to-end business processes” and move organizations “from manual tasks to autonomous, insight-driven operations.”
The distinction matters because the two demand different governance. Copilot inherits the calling user’s permissions; an agent needs its own identity, its own boundaries, and its own review cadence. Treating an agent like a Copilot feature (switch it on, see what happens) is how organizations discover that AI-added thinking applied to AI-first tooling produces neither. Our overview of agentic AI solutions for Dynamics 365 covers this architectural distinction in depth.
The 7 Shifts the 2026 Agent Lineup Introduces
1. Order capture stops being a typing job.
Conventional pattern: Customer emails arrive in a shared inbox; coordinators interpret, rekey, and validate each one manually. This is the single largest clerical load in most distribution back offices.
AI-first replacement: The Sales Order Agent monitors the inbox, extracts items and quantities, validates against live inventory, flags shortages, and drafts the order for one-click human review. From July 2026, companies can run multiple Sales Order Agents per company with smarter item matching and relevance filtering, so different inboxes, brands, or customer segments get their own agent.
Operational impact: Removes 40–60% of order-desk keying effort and cuts email-to-order latency from hours to minutes.
2. AP matching becomes an exception process.
Conventional pattern: Clerks capture vendor invoices, match them to purchase orders and receipts three ways, and chase discrepancies, most of it on transactions that were never going to fail.
AI-first replacement: The now generally available payables The agent captures invoices in multiple languages, runs 3-way matching, categorizes processed emails in the shared Outlook mailbox so nothing is double-handled, and fast-tracks trusted vendors through Known Senders (August 2026). Humans review exceptions rather than the whole population.
Operational impact: Recovers 15–25% of AP processing time and surfaces discrepancies at capture rather than at month-end.
3. Expense processing joins the agent layer.
Conventional pattern: Employees hoard receipts, submit spreadsheets at month-end, and finance re-keys, categorize, and chase policy exceptions weeks after the spend occurred.
AI-first replacement: The Expense Agent (public preview May 2026, mobile capture from July 2026) lets employees photograph receipts on their phone, then categorizes and itemizes the expense with AI and routes it through approval workflows, with project assignment and travel requisitions on the roadmap through September 2026.
Operational impact: Collapses the submit-to-post cycle from weeks to days and removes the finance re-keying layer entirely for compliant expenses.
4. Security gains a new class of principal and a control room.
Conventional pattern: Role-based security was designed for humans: one person, one role, one accountability chain.
AI-first replacement: Each agent carries its own permission sets and a dedicated profile constraining which pages and actions it can even see. The 2026 Wave 1 release adds the operational controls that make this manageable at scale: a dedicated task pane for tasks from all agents, agent-generated content reviewable directly on documents, created-by and modified-by AI indicators on records, and the ability to stop all active tasks for a selected agent instantly.
Operational impact: Agent activity is fully attributable, reviewable in context, and haltable in one action, a precondition for SOC 2 and ISO environments, not a nice-to-have.
5. The audit trail becomes step-level, not record-level.
Conventional pattern: The change log tells you a field changed and who was logged in; reconstructing why requires archaeology.
AI-first replacement: Agent task logs record every page the agent visited, every decision it made, every escalation it raised, and the cost of the run, and the 2026 wave 1 MCP enhancements extend this with telemetry and audit logging across agent access paths. New APIs for analyzing approval workflows and permissions (June 2026) give auditors direct programmatic access.
Operational impact: Audit preparation for agent-touched processes becomes a query, not a project, and reviewers see reasoning, not just outcomes.
6. Automation acquires a metered cost line.
Conventional pattern: Automation cost was capitalized into implementation: build once, run free.
AI-first replacement: Built-in and custom agents consume Copilot credits per task, provisioned prepaid or pay-as-you-go, and Microsoft is lowering the trial barrier (Payables Agent free trial without email setup, July 2026) precisely because it expects consumption to become a standing line item.
Operational impact: Agent economics become a live governance question (which processes justify per-run AI cost versus deterministic workflow), and organizations that skip this analysis find the answer on the invoice.
7. Business Central opens to the agent ecosystem via MCP.
Conventional pattern: External integrations meant OData feeds and middleware, consumed by systems, not reasoning agents.
AI-first replacement: The MCP server, introduced in the prior wave, gains substantial 2026 wave 1 enhancements: broader data access (June 2026), agentic experience support, and a connection from AI agents into the Business Central admin center itself (April 2026), so agents can participate in environment administration, not just transactions. Custom agents built in Copilot Studio or AL plug into the same permission and audit boundary.
Operational impact: Business Central becomes addressable by the broader agent ecosystem, which widens both the capability surface and the governance surface at the same time.
Taken together, these shifts do to routine transaction handling what we described in reducing manual work in Business Central without customization, except the actor doing the work is no longer always a person.
Copilot-Assisted Business Central vs. Agent-Operated Business Central
How AI in Business Central splits into Copilot-assisted and agent-operated architectures (2026 wave 1 state)
|
Trigger |
User asks |
Event, email, receipt, schedule, or task |
|
Execution |
Single response, user acts |
Autonomous multi-step background |
|
Identity |
Inherits the calling user |
Own profile + scoped permission sets |
|
Human role |
Actor for every step |
Approver at defined thresholds, via cross-agent task pane |
|
Visibility |
Chat history |
Task logs, on-page content review, created-by-AI indicators |
|
Emergency control |
Close the pane. |
Stop all active tasks per agent instantly. |
|
Cost model |
Included in licensing tiers |
Copilot credits per task |
|
Extensibility |
Fixed Microsoft features |
Custom agents via AL, Copilot Studio, and MCP |
|
Failure mode |
Bad suggestion; a human catches it. |
Bad instruction, agent repeats it; governance required |
The right-hand column is where Microsoft is taking the product, and the 2026 wave 1 additions are all on that side of the table. The relevant question is not whether your Business Central tenant will have agents (the release cadence guarantees it will) but whether they arrive into an environment designed for them.
Prerequisites Before You Activate Agents
-
- Clean master data. An agent validating orders against inventory inherits every duplicate item, stale unit of measure, and unblocked obsolete SKU. Data quality problems become agent behavior problems at machine speed.
- Documented process boundaries. Agents need explicit thresholds (order values, match tolerances, credit limits, expense policy caps) that most organizations hold as tribal knowledge. Write them down before writing agent instructions.
- A permission model designed for non-human principals. Least-privilege permission sets and agent-specific profiles must exist before activation, not after the first incident. Use the new permissions audit APIs (June 2026) to verify.
- A credit governance policy. Decide which processes justify per-task Copilot credit consumption, set consumption monitoring, and assign ownership of the AI cost line.
- An exception-handling owner. Every agent escalation needs a named human with a response SLA and the cross-agent task pane as their working queue; otherwise, human-in-the-loop becomes human-in-the-queue.
- Update-cadence awareness. Agent capabilities ship incrementally within each wave (Expense Agent capabilities landed in May, July, and September 2026 separately). Someone must own the release plan and review each wave. 1.
The Cost of Watching This Wave From the Sidelines
The defer-and-observe position sounds prudent: let the previews mature, and revisit in a year. Its cost is structural rather than dramatic. Every month of deferral is another month of orderdesk, AP, and expense-processing capacity spent on work the platform can now execute, labor that, in a mid-market operation, compounds to 1,000+ hours annually across order capture and invoice matching alone. That capacity is not recovered retroactively.
The subtler cost is organizational readiness. Deploying agents well requires clean data, documented thresholds, and a governance model: assets that take months to build and cannot be purchased at go-live. Organizations that start now build them against low-stakes, human-reviewed agent workloads, with the Payables Agent free trial as a zero-commitment proving ground. Organizations that defer will build them under pressure when future waves make agents the default posture of the product rather than an opt-in. With the governance controls now shipped, 2026 is the cheapest moment there will ever be to learn this discipline. H
How Omni Logic Solutions Delivers:
The AI-First Framework
Activating agents is a settings page. Deploying them safely is a methodology. Omni Logic Solutions delivers AI in Business Central through its AI- First Framework:
-
- AI-First Discovery: Workflows classified as human, Copilot-assisted, or agent-automated before config.
- Copilot-Grounded Data Design: Taxonomy, security, and retrieval boundaries to eliminate hallucination.
- Human + Agent Workflow Build: Approval thresholds, exception handling, and handoffs before agent config.
- Pre-Go-Live Agent Governance: SOC 2/ISO audit trails, role-scoped permissions, and Copilot Studio guardrails.
- AI Adoption Programme: Change management for working alongside agents, not ERP feature training.
This is the “Built AI-First, Not Bolted-On” position in practice: the agent boundary is designed at project initiation, not retrofitted after the first ungoverned escalation. Initial agent deployments typically run ~90 days; broader multi-process rollouts run 4–9 months depending on data readiness.
Closing Perspective
The 2025 wave 2 release introduced agents; the 2026 wave 1 release made them dependable. The Payables Agent at GA, the Expense Agent in preview, and multi-agent order capture are the visible artifacts. The deeper change is architectural: a platform where non-human principals hold permissions, execute work, leave step-level audit trails, and can be reviewed, attributed, and stopped from a single pane, extensible by anyone who can write clear instructions.
The adopt-now case is not about novelty. Early adopters get to build agent governance against forgiving, human-reviewed workloads and bank the recovered capacity immediately.
Built AI-First, Not Bolted-OnDeferrers will adopt the same capabilities later (Microsoft’s release cadence assures that) but under conditions they did not choose, with data they did not clean, at a moment the update calendar picks for them. The technology will wait. The advantage will not.
Not Sure If Your NAV Environment Is Migration-Ready?
Book a working session with Omni Logic Solutions. We will walk your version, ISV stack, and customization footprint and give you a scored, no-fluff readout.
FAQs
What are AI agents in Business Central?
Agents are autonomous, AI-powered actors that operate inside the tenant under their own identity. Unlike Copilot, which responds to user requests, agents are triggered by events (emails, receipts, transactions, schedules) and execute multi-step tasks within scoped permissions, escalating to humans at defined thresholds. See our agentic AI solutions for Dynamics 365 overview for the architecture.
What agents are available in Business Central in 2026?
Three first-party agents: the Sales Order Agent (email-to-order capture, now supporting multiple agents per company), the Payables Agent (generally available, with 3-way matching, multi-language capture, and Known Senders), and the Expense Agent (public preview from May 2026, with mobile receipt capture from July). Custom agents can be built via AL, Copilot Studio, and the MCP server, per Microsoft’s 2026 wave 1 release plan.
How is an agent different from Copilot in Business Central?
Copilot is interactive: a user asks, it responds, and the user acts. An agent is autonomous: it interprets a goal, plans steps, navigates Business Central pages under its own identity and permissions, and involves humans only where its instructions require. The governance surface (separate profiles, permission sets, task logs, on-page content review, and an instant stop control) is the practical difference.
What do Business Central agents cost to run?
Built-in and custom agents consume Copilot credits per task, provisioned through prepaid or pay-as-you-go models. This makes agent economics a governance question: high-volume, rules-only processes may still belong in deterministic workflows, while judgment-adjacent processes justify per-run AI cost. The Payables Agent free trial (July 2026) lets you measure real consumption before committing.
Can we build our own agents without heavy development?
Largely, yes. Custom agents can be defined with natural-language responsibilities, guidelines, and instructions, with control keywords for hard stops and human escalation, and Copilot Studio agents connect to Business Central through the enhanced MCP server. Production-grade agents can then be graduated into AL, where the 2026 wave 1 tooling adds agent session debugging and AI evaluation support.
Do agents replace ERP staff?
No. The current agents absorb the transactional layer (re-keying, matching, validating, and expense categorization), while humans retain approvals, exceptions, and judgment. The practical effect is that transaction volume decouples from headcount, which changes hiring plans rather than eliminating roles.
What should we fix before switching agents on?
Master data quality, documented process thresholds, a least-privilege permission model for non-human principals, and a Copilot credit governance policy. Agents amplify whatever environment they are deployed into, including its defects. Our guide to Business Central agent-driven operations covers the readiness sequence.
Is AI in Business Central relevant if we only use it for finance?
Yes. The Payables Agent, the Expense Agent, and the close acceleration they enable are finance-first capabilities. Paired with a disciplined month-end closing process in Business Central, agent-validated sub-ledgers materially shorten the close.
-
Order Fulfillment Process: 7 AI-First Controls That Protect Commitments
Microsoft Dynamics 365 Business Central Edition Insights, updates, and practical guides for Microsoft Dynamics 365 Business Central Home Order Fulfillment Process: 7 AI-First Controls That Protect Commitments Vishal R October 7, 2026 1:56 pm Most distributors can describe their order fulfillment process: capture the order, confirm supply, allocate inventory, release work, ship, invoice, and update
October 7, 2026 -
Power BI Consultant: 10 Controls for Operational Dashboards
Microsoft Dynamics 365 Business Central Edition Insights, updates, and practical guides for Microsoft Dynamics 365 Business Central Home Power BI Consultant: 10 Controls That Make Dashboards Operational Vishal R September 25, 2026 4:22 pm Most organizations do not have a dashboard shortage. They have a decision-quality shortage. Finance, operations, sales, and customer service can each
September 25, 2026 -
ERP and Accounting Systems: When to Upgrade From Your Accounting Software?
The ERP & CRM Knowledge Hub Strategy, Selection & Comparison Guides For Canadian Businesses Home ERP and Accounting Systems: When to Upgrade From Your Accounting Software? Vishal R September 21, 2026 11:20 am Most growing businesses reach the ERP-versus-accounting-software question the same way: not by planning for it, but by accumulating workarounds until someone finally
September 21, 2026 -
AI in Business Central 2026: The Agents Changing Your ERP
Microsoft Dynamics 365 Business Central Edition Insights, updates, and practical guides for Microsoft Dynamics 365 Business Central Home AI in Business Central 2026: The Agents Changing Your ERP Vishal R September 17, 2026 9:51 am For twenty years, the operating assumption behind every Dynamics 365 Business Central deployment, and Navision before it, was that the
September 17, 2026 -
Top Accounting Software for Canadian Small Business : Comparison Guide (2026)
The ERP & CRM Knowledge Hub Strategy, Selection & Comparison Guides For Canadian Businesses Home Choosing The Best Accounting Software for Canadian Small Business in 2026 Vishal R August 18, 2026 1:12 pm Most Canadian small businesses choose their accounting software once, early, and under pressure a founder signs up for QuickBooks or Wave in year one because the
August 18, 2026 -
Logistics vs Supply Chain Management: The D365 Difference
Logistics is a subset of supply chain management. Logistics moves and stores goods. Supply chain management plans, sources, and orchestrates the entire flow, then hands physical execution to logistics.
August 7, 2026
