The phrase “agentic ERP” has been making the rounds since March. Microsoft put it in their own blog: Business Central “accelerates the move to agentic ERP with enhancements to our AI-powered agents that automate sales and purchase scenarios” (Microsoft Dynamics 365 Blog, March 18, 2026).
That marketing line cashes out in concrete software. As of Update 28.0, which started rolling out on May 8, 2026, Business Central includes three production-ready AI agents (Payables, Sales Order, and Agent Designer for custom builds), one in public preview (Expense Agent), a benchmarking framework for evaluating AI-built code (BC-Bench), and a mandatory MCP server layer that lets external AI clients connect to BC. There is a lot to unpack.
This guide is the reference piece. It walks through every agent and every supporting capability that landed in Wave 1: what each does, how it works under the hood, what setting it up looks like, where the documented limitations are, who it is designed for, and what the cost model is. Read it as the depth source.
For the controller’s decision framework on whether and when to turn each agent on, see the companion post: Before You Turn On Business Central’s AI Agents: A Controller’s Pre-Flight Checklist.
How Business Central AI Agents Actually Work
Before getting into specific agents, a quick word on what makes them different from the Copilot features that have already been in Business Central for the last 18 months. The distinction matters more than the marketing makes obvious.
Copilot is a conversational assistant. It has been in BC for a while. You ask, it responds, you act. The user stays in the driver’s seat for every step.
An agent is autonomous within bounded scope. A BC agent operates in the background, watches for trigger conditions (an inbound email, a new request), performs multi-step work, and routes the result to a human for approval. It does not wait for a prompt. Microsoft frames the shift in the Wave 1 overview: Business Central is moving “from manual tasks to autonomous, insight-driven operations” and increasingly “acts as an active partner in daily work” (Microsoft Learn, Wave 1 plan).
The architecture is layered, and Microsoft is explicit about the layering in the Agent Designer documentation:
“Business Central agentic strategy intentionally separates deep, well controlled, in-context and transparent execution within the product from cross-system orchestration. Business Central agents form the foundational layer inside the product, exposing well-defined agent capabilities that in the future will be possible to surface as MCP tools and consume by Declarative Agents in Microsoft Copilot Studio and Microsoft 365.”
In practical terms:
- BC agents are the in-product layer. They run inside Business Central, operate against BC data, and use the same security and permission model as a human user. Every action they take is logged and attributed to the agent identity. If you can audit what a user did, you can audit what an agent did.
- MCP server is the connectivity layer. Model Context Protocol is the standardized way external AI clients (Visual Studio Code, Copilot Studio, custom internal tools, third-party agents) can call into BC’s data and capabilities. In Update 28.0, MCP server access is mandatory and cannot be turned off (Microsoft Learn, BC 28.0 what’s new). The infrastructure for external AI agents to interact with BC is now always on.
- Copilot Studio is the cross-system orchestration layer. Agents that span multiple Microsoft products (BC + Outlook + Teams + SharePoint, for example) are built and managed in Copilot Studio, which calls BC agents through MCP.
That layering matters for two reasons. First, it tells you where customization actually happens. A BC-only agent goes into Business Central via Agent Designer. A cross-product agent goes into Copilot Studio and calls BC. Second, it tells you where to look for governance: BC agents are governed by BC’s permission model; Copilot Studio agents are governed by Copilot Studio’s separate model. Confusing the two is how organizations end up with audit gaps.
One constraint applies to every agent below: Copilot and agents are available only to Business Central online (SaaS) customers (Microsoft Learn, Wave 1 plan). On-premises and partner-hosted BC deployments do not have access to agent features. If you are still on legacy NAV or on-prem BC, the question is a migration question first.
The Payables Agent
The Payables Agent automates the front end of accounts payable: receiving vendor invoices, reading them, locating the right vendor and accounts in Business Central, and drafting an invoice entry for human review.
What it does
From the official Payables Agent documentation:
“The Payables Agent monitors mailboxes for incoming vendor invoices, uses AI to analyze invoice content, and shows invoice drafts to agent supervisors for review. Ideally, users make few or no corrections before finalizing drafts, so invoices are ready for approval and posting.”
The stated goal is plain-spoken: “get accounting skills and help register invoices correctly, ideally removing bottlenecks in accounts payable so finance support doesn’t slow company growth.”
The workflow goes like this. A vendor sends an invoice PDF as an email attachment to a mailbox the agent is watching. The agent reads the PDF (Microsoft’s documentation indicates Azure Document Intelligence under the hood). It identifies the vendor in the BC vendor master, picks the appropriate purchase account, and extracts header and line detail. It creates a draft purchase invoice in BC and surfaces it to a designated supervisor. The supervisor reviews, corrects if needed, and finalizes for approval and posting through whatever AP process the company already runs.
How it works under the hood
The agent runs autonomously in the background on a schedule. It uses Microsoft’s AI services to perform OCR and entity recognition on the PDF, then maps extracted entities to BC master data using essentially the same logic an experienced AP clerk would: vendor name match, address match, tax ID match, account code in the line description.
Confidence scoring is part of the design. When the agent’s match confidence is low (an unfamiliar vendor, an ambiguous line item, an unusual amount), the draft is flagged for closer supervisor review rather than presented as ready-to-post. This is what a controller would build by hand if they could, and the agent does it by default.
Documented limitations
This is the section most BC partner blog posts skip. Read the documentation and the constraints are listed plainly:
- The agent only processes emails with PDF attachments
- Skips emails with more than 10 attachments
- Does not process PDFs larger than 5 MB
- Does not process PDFs with more than 10 pages
- Processes a maximum of 100 emails per day
- Processes a maximum of 500 invoices per day
The current feature gaps, also verbatim from Microsoft: “The Payables Agent currently doesn’t support the following features: Purchase order matching / Approval flows / Anomaly detection.”
And a language note: “This Copilot feature is validated and supported in English only. While it can be used in other languages, it might not function as intended.”
Take a minute on these. A $20M services firm with 30 vendor invoices a day, English-only AP, no three-way-match requirement, and PDF-only vendor invoices is squarely in the Payables Agent’s sweet spot. A $200M distributor with 800 invoices a day, multi-language vendor base, mandatory three-way-match, and a mix of PDF and EDI invoices is going to design around several of these limits or wait for Wave 2.
Availability
The Payables Agent is generally available in all countries and regions per the Copilot and Agents availability table, with the English-only validation caveat noted above.
Who it is designed for
In practice, the Payables Agent fits best when:
- The organization runs BC online (not on-premises)
- Vendor invoices arrive primarily as PDF email attachments
- The AP workflow is English-only (or willing to pilot in the English-language vendor segment first)
- Three-way matching to a purchase order is not a hard requirement today
- Daily invoice volume is comfortably under 500 per BC environment
- A designated AP supervisor will review agent-drafted invoices before final posting
The Sales Order Agent
The Sales Order Agent handles the front of order-to-cash. Customers email in requests, the agent interprets the request, validates customer and product details against BC, calculates availability, and drafts a sales quote.
What it does
From the Sales Order Agent documentation:
“Sales Order Agent helps Business Central users automate the process of capturing sales orders. The agent uses AI to analyze customer requests received via email and locates the customer in Business Central. If important details are missing or more choices are available, the agent engages in multi-turn email conversations to clarify the request. It checks and informs the customer about the availability of the items they’re looking for. The agent then follows up with a sales quote.”
The keyword in that description is multi-turn. The Sales Order Agent does not just parse a single email; it has a conversation. If the customer’s request is missing a delivery date, the agent emails back asking. If the customer specifies a product family but not a specific SKU, the agent emails back with options. That is much closer to how a real inside-sales rep handles request emails than how traditional order-entry automation works.
The CTP integration
The detail that gives this agent operational teeth, especially for manufacturers and distributors, is its integration with Capable-to-Promise (CTP). From the docs:
“The Sales Order Agent includes an option to use Business Central’s capable-to-promise (CTP) functionality that lets the agent calculate the earliest possible delivery date for items not currently in stock. CTP evaluates production capacity, procurement timelines, and supply chain constraints to determine when an item can realistically be delivered.”
In an organization with mature BC manufacturing or warehousing setup, this means the agent can give a defensible delivery date for an out-of-stock item without anyone manually involving production planning. The agent walks the BOM, looks at component availability, considers production capacity and procurement lead times, and produces a date the company can actually meet.
For a manufacturing or distribution operation where availability and lead-time questions are the bottleneck on quoting today, this is the agent that pays the bill.
Human approval is required
Every outbound message from the Sales Order Agent goes through a human review gate. The documentation is direct: “The agent always involves designated Business Central users to review and approve all outgoing messages before they’re sent to the customers.”
That is by design. The agent drafts the response, the assigned reviewer approves it, and only then does it go to the customer. So the agent does not “automate sales orders end-to-end” in the sense some marketing copy implies. It substantially reduces time per quote, but a human still owns the customer communication.
Availability and language support
The Sales Order Agent is generally available in all countries and regions. Language coverage is broader than the Payables Agent’s: Danish, Dutch, English (Australian, Canadian, British, New Zealand, and US variants), Finnish, French, German, Italian, Norwegian Bokmål, Spanish, and Swedish are all on the availability table.
Who it is designed for
The Sales Order Agent fits best when:
- Inbound customer order requests arrive primarily by email (rather than EDI or B2B portal)
- The business is manufacturing, distribution, or made-to-order, where availability and delivery-date questions slow down quoting
- CTP or basic Available-to-Promise (ATP) is configured in BC
- Inside-sales or order-management staff can act as agent reviewers
- Faster quote response matters more than full end-to-end automation
The Expense Agent
The Expense Agent is the newest of the named agents. Microsoft’s Aleksandar Totovic, Principal Product Manager, announced it on April 27, 2026, with public preview starting in May. It captures receipts, categorizes expenses, handles mileage and per diem, routes project expenses to the right ledgers, and reinvoices billable expenses automatically.
What makes it different from the other agents
Two characteristics make the Expense Agent a meaningful departure from how BC has handled this category historically.
It is designed for users who do not have a Business Central license. Microsoft states it explicitly: “No Business Central licenses required.” The agent’s target users are field technicians, drivers, traveling salespeople, project consultants. People who incur expenses but never log into BC themselves. They interact through a dedicated web app, Outlook, or Microsoft 365 Copilot Chat. The expense data flows into BC; the user never sees the BC interface.
For organizations whose field workforce historically used external expense management tools, this changes the build-or-buy conversation. You now have a native option that integrates directly into BC without seat costs for the field users.
Usage is metered through Copilot Credits, not seat licenses. This is the same billing model used by the Payables and Sales Order agents (see Section 7 below), but it is the first time it has been used to extend BC capabilities to a non-BC-licensed population. The financial trade-off is no longer “is the per-seat cost worth it for an employee who only files expenses?” It is “what does Copilot Credit consumption look like at our expense-report volume?”
Documented limitations
At public preview launch:
- US only at launch
- English only at launch
- Additional languages and regions expected starting July 2026, per the announcement post
This is a meaningful constraint for any organization with international employees expensing in non-English receipts. The Expense Agent today is a US-pilot conversation, not a global rollout.
Who it is designed for
The Expense Agent fits best when:
- The organization runs BC online and operates primarily in the US (at preview)
- A meaningful population of non-BC-licensed employees incurs expenses
- The current expense management tool’s contract is up for renewal
- Native BC integration for expense data is preferred over a third-party system
- Project expense reinvoicing or mileage tracking is a material part of the workload
Agent Designer and Custom Agents
For organizations whose needs do not fit neatly into Payables, Sales Order, or Expense, Agent Designer is the path to custom-built BC agents.
What it is
Agent Designer reached general availability in May 2026 after a public preview that began February 6, 2026.
Microsoft’s positioning is broader than just developers:
“This approach empowers partner personas like consultants, product owners, and domain experts, as well as developers and even power users, to explore new scenarios, validate ideas, and build confidence in their agents before moving to production.”
In plain terms: Agent Designer lets people who are not full-stack AL developers prototype custom agents using natural-language instructions. It is positioned to be accessible to a consultant or a domain expert in finance or operations, not just to BC ISVs and the AL development community.
How it works
A custom agent built in Agent Designer extends the same runtime as the built-in Payables and Sales Order agents:
“The experience offers a low-risk environment for envisioning and prototyping new AI agents tailored to your own business scenarios, extending the same agent runtime capabilities used by built-in Business Central agents.”
The author of the agent specifies what the agent should do in natural language, points it at the BC data and capabilities it needs access to, and tests it in a sandbox environment. Permissions, audit logs, and security inherit from BC’s standard model. A custom agent does not get any access that a human user with the same permissions would not have.
Sandbox to production: Update 28.1
The sandbox-to-production path closes in Update 28.1, which Microsoft MVP Stefano Demiliani has documented is rolling out May to June 2026: “Dynamics 365 Business Central update 28.1 will allow you to deploy your own Business Central agents to production environments” (Demiliani, May 10, 2026).
Before 28.1, Agent Designer is a prototyping tool. After 28.1, custom agents can be deployed to production environments and run alongside the built-in agents.
Examples of what fits
Microsoft has not published a definitive list of custom-agent use cases. From the design (agents that operate inside BC, on BC data, with BC’s security model) reasonable first builds include:
- An industry-specific Payables variant that handles vendor invoices the built-in Payables Agent cannot. For example, a vendor that sends invoices as images embedded in the email body rather than as a PDF attachment
- A statement-reconciliation agent that matches incoming bank or vendor statements against BC ledgers and flags discrepancies for review
- A customer-correspondence agent that responds to status inquiries on order shipments, drawing from BC sales order and shipment data
- A new-vendor-setup agent that gathers vendor onboarding information through a structured email exchange and prepares the BC vendor card for approval
Who it is designed for
A custom agent makes sense when:
- A specific recurring workflow does not fit one of the built-in agents but does fit BC’s agentic model (event-triggered, multi-step, ending in a human-approved BC action)
- The organization has a consultant or business analyst who can author the agent in natural language, or the BC partner does
- The cost of the workflow today is large enough to justify the design and ongoing Copilot Credit consumption
- The workflow does not require crossing system boundaries (cross-system agents belong in Copilot Studio, not Agent Designer)
Supporting Infrastructure: MCP Server, BC-Bench, and AL Runner
Several supporting capabilities shipped in Wave 1 alongside the named agents. None are agents themselves, but each shapes how agents are built, tested, and connected.
MCP server
The Model Context Protocol (MCP) server in BC 28.0 is the connectivity layer that lets external AI clients call into Business Central. As noted earlier, it is mandatory in 28.0. The feature is enabled for all users and disabling it has no effect (Microsoft Learn, BC 28.0 what’s new).
What this means practically: any MCP-compliant AI client (Visual Studio Code with appropriate extensions, Copilot Studio, Claude Desktop, custom internal tools) can connect to BC and use BC’s data and capabilities as tools. The same security and permission model that governs human users governs MCP connections.
For most CFO and controller audiences, MCP is infrastructure they will hear about but not directly configure. For IT and partner audiences, MCP is the long-term direction for how all AI tooling will connect to BC. Treat it as the equivalent of an API gateway purpose-built for AI clients.
BC-Bench
BC-Bench is a Microsoft-built benchmarking framework for evaluating AI agents on real-world Business Central AL coding tasks. Community blogger Ian Grieve documented its purpose in detail: “a benchmarking framework for evaluating agent performance on real-world Dynamics 365 Business Central AL coding tasks” and measuring “realistic Business Central development scenarios, such as bug fixes and test creation, using curated AL code problems derived from real pull requests” (azurecurve, March 30, 2026).
BC-Bench is a developer-facing tool. Its relevance to a finance or operations reader is indirect but real: it signals that Microsoft is building the quality-bar infrastructure to evaluate whether AI-generated BC customizations actually work before they reach production. If you eventually approve a custom agent built with Agent Designer, the development work behind it is increasingly likely to have been benchmarked through this framework.
AL Runner
AL Runner is an independent open-source tool from Microsoft MVP Stefan Maron. It is not a Microsoft product, but it has become part of the Wave 1 conversation because of what it enables. From Maron’s documentation: “the BC compiler, when used directly, exposes public API to transpile AL into C# representation”, which lets developers run AL unit tests in seconds without spinning up a BC service tier or Docker container.
For AI coding agents that are writing and testing AL code, AL Runner removes 15 to 30 minutes of pipeline overhead per test cycle. This is one of the reasons Microsoft’s own benchmarking framework (BC-Bench) is becoming practical to run at scale.
Again, this is developer infrastructure. The signal for a business audience is the same as BC-Bench: the toolchain for building and testing AI-driven BC customizations is being rebuilt to make agentic development faster and more reliable.
The Copilot Credit Cost Model
All three named agents (Payables, Sales Order, Expense) and custom agents built with Agent Designer consume Copilot Credits, not flat per-user or per-transaction licenses. From the Agent Designer documentation:
“custom agents consume Copilot credits as they perform their steps. Credits can be provisioned through prepaid or pay-as-you-go models, providing flexible cost management as your AI usage scales.”
How consumption varies
Microsoft MVP Stefano Demiliani has documented that credit consumption “depends on the feature used and the actions that your agent performs” (Demiliani, March 24, 2026). Different agent actions consume different credit amounts. A simple Payables Agent draft of a clean invoice and a multi-turn Sales Order Agent conversation with CTP lookups will not consume the same credits.
So your monthly Copilot Credit cost is variable, driven by your actual transaction mix and volume. You are building a new cost line in your finance budget that operates differently from per-seat BC licensing.
Prepaid versus pay-as-you-go
Microsoft offers both billing approaches.
- Prepaid Copilot Credits give you predictable budget control. You buy a block of credits, the agents consume them, and when the block is exhausted, agent operations pause until you provision more. The failure mode is service interruption, but the financial exposure is bounded.
- Pay-as-you-go Copilot Credits give you elastic capacity. The agents consume credits without a hard cap, and you get an invoice for actual consumption. The failure mode is an unbounded invoice if consumption spikes, but service continuity is uninterrupted.
Neither model is the right answer for every organization. A finance team that prioritizes budget predictability typically chooses prepaid. A finance team that prioritizes operational continuity typically chooses pay-as-you-go.
Cost visibility today
Demiliani also documents the permission model for monitoring consumption inside BC: “Only SUPER and AGENT – DIAGNOSTICS permission sets can view billing details; AGENT – ADMIN permissions are insufficient.” If you want a controller or AP supervisor to monitor agent consumption inside BC, the standard AGENT-ADMIN permission set is not enough. The right permission must be assigned deliberately.
What Microsoft does not publish
Microsoft does not currently publish a per-action or per-invoice credit cost. There is no flat “$X per Payables Agent invoice” or “$Y per Sales Order Agent quote” rate. The way to project monthly cost today is to run a one-week pilot on a controlled subset of transactions, measure actual credit consumption, and extrapolate. Anything else is guesswork dressed up.
Permissions and Governance
The BC agent permission model is built on the same security framework that governs human users in Business Central, with some agent-specific additions.
Standard agent permissions
Microsoft documents two primary agent-related permission sets in Wave 1:
- AGENT – ADMIN: Standard administration of agent configuration. Can configure which agents are active, what mailboxes they monitor, and which users are designated reviewers.
- AGENT – DIAGNOSTICS: Visibility into agent operational data, including Copilot Credit consumption details. Required for cost monitoring.
A user with AGENT – ADMIN can manage what the agents do but cannot see what the agents are costing without AGENT – DIAGNOSTICS or SUPER permissions. That separation is deliberate. Managing operations and monitoring spend are different jobs that should be assigned to different people.
Built-in governance affordances
Several Wave 1 features address agent governance explicitly:
- “Manage tasks from all agents in dedicated task pane” (GA April 1, 2026): gives administrators and reviewers a single place to monitor agent activity across all active agents
- “Stop all active tasks for selected agent” (GA April 1, 2026): emergency-stop affordance for any agent
- “Review content generated by agents directly on pages” (GA April 1, 2026): supervisor review workflow integrated into the BC pages where the affected records live
- “Discover emails in the mailbox that have been processed by Payables Agent” (GA April 1, 2026): audit visibility into what the Payables Agent has handled
These are the operational governance controls Microsoft shipped explicitly because agents represent a different kind of trust relationship than traditional automation. Agents take action. Humans review. Administrators have visibility and emergency override. That triad is the architecture.
What governance still requires you to design
The agent permission model handles authorization and audit, but several governance questions remain organization-specific:
- Who is the designated reviewer for each agent, and what is their queue capacity?
- What is the exception path when an agent surfaces something the reviewer is not sure about?
- What is the change-management process for activating, deactivating, or modifying an agent?
- How are agent costs allocated to departments or cost centers in your internal financial reporting?
None of these are answered by Microsoft’s documentation. They are policy decisions that belong to the finance, IT, and operations leadership of the organization deploying the agents. If you are the one being asked to turn on the Payables Agent, you are also the one being asked to answer these four questions. Have answers ready.
What is Available
A summary of agent and agent-adjacent feature status as of May 11, 2026.
Generally available (Wave 1, BC 28.0):
- Payables Agent (all countries/regions, English-only validation)
- Sales Order Agent (all countries/regions, 11 supported languages)
- Agent Designer (custom agent prototyping)
- MCP server access (mandatory in BC 28.0)
- Dedicated agent task pane
- Review agent-generated content on pages
- Stop all active tasks for an agent
- BC-Bench (AL coding agent benchmarking)
- Discover Payables Agent processed emails
In public preview:
- Expense Agent (US-only, English-only at launch; broader rollout starting July 2026)
- Connect AI agents to admin center through MCP server (preview as of Update 28.0)
Coming in Update 28.1 (May–June 2026):
- Custom agent deployment to production environments (per Demiliani’s MVP coverage)
- GPT-5.3-chat as the new default model for agents (rolling out across regions)
- UI-based model selection inside the Agents page
Not yet supported in current Payables Agent (per Microsoft documentation):
- Purchase order matching
- Approval flows
- Anomaly detection
These last three items are explicit gaps in the current Payables Agent. Microsoft has not committed to a specific Wave or version for adding them, but their explicit listing in the limitations section is a signal that they are on the product roadmap. If you decide to wait on the Payables Agent, this is what you are waiting for.
Where to Go From Here
This guide is the reference piece for what Business Central’s AI agents are and how they work. The next decision (whether and when to enable each agent for your organization) is a different question and depends on your specific finance and operations workflow.
For the controller-grade framework on that decision, including the pre-flight checklist for evaluating fit before turning any of this on, see the companion post: Before You Turn On Business Central’s AI Agents: A Controller’s Pre-Flight Checklist.
For broader context on AI strategy across the Microsoft stack, our Keys to AI Success With Microsoft Copilot guide covers the organizational framing. For grounding in how Business Central itself is structured beneath the agent layer, the Complete Guide to Business Central Modules lays out the platform the agents operate on top of.
When you are ready to pilot a specific agent against your actual transaction volume, or to design a custom agent that solves something the built-in pair does not, Tigunia’s Business Central practice can help structure the pilot and translate the operational results into the financial framing your CFO needs.
FAQS: Guide to Business Central AI Agents
What is agentic ERP in Business Central?
Microsoft’s 2026 Release Wave 1 framing describes a shift in Business Central from manual tasks to “autonomous, insight-driven operations” through AI agents that operate inside BC using the same security and permission model as human users. As of May 2026, the generally available agents are the Payables Agent and Sales Order Agent (since April 2026) and Agent Designer for custom agents (since May 2026). The Expense Agent is in public preview.
What is the difference between Copilot and an AI agent in Business Central?
Copilot is a conversational assistant that responds to user prompts. An agent operates autonomously in the background, triggered by events (an inbound email, for example), executes multi-step work, and routes results to a human reviewer for approval. Copilot waits for you; agents watch for trigger conditions and act on their own within their defined scope.
Are AI agents available on Business Central on-premises?
No. Microsoft’s Wave 1 overview states explicitly that Copilot and agents are available only to Business Central online (SaaS) customers. On-premises and partner-hosted BC deployments do not have access to these capabilities.
What does the Payables Agent cost?
Built-in agents (Payables, Sales Order) and custom agents built with Agent Designer consume Copilot Credits per task. Credits are available as prepaid or pay-as-you-go. Microsoft does not publish a flat per-invoice price; consumption varies by agent feature and actions performed. The recommended way to project monthly cost is a one-week pilot on representative volume.
How does the Sales Order Agent calculate delivery dates?
When configured with Capable-to-Promise (CTP) integration, the Sales Order Agent evaluates production capacity, procurement timelines, and supply chain constraints inside BC to determine the earliest realistic delivery date for out-of-stock items. This uses the same CTP logic available to BC users today; the agent automates the calculation as part of the quote-drafting process.
Can a custom AI agent built with Agent Designer access any BC data?
A custom agent inherits the permissions of the user account it runs under. It can access the data and capabilities that user account is permitted to access, no more and no less. Microsoft’s Agent Designer documentation states that every action is logged and permissions are enforced.
When did Business Central 2026 Wave 1 (Update 28.0) release?
Update 28.0 became available to existing customers with environments scheduled to update beginning May 8, 2026. Administrators can schedule the 28.0 update to any date within the 5-month update period, which ends August 31, 2026.
What is BC-Bench?
BC-Bench is a Microsoft-built benchmarking framework for evaluating AI agent performance on real-world Business Central AL coding tasks. It uses curated AL problems derived from real pull requests to measure how well AI coding agents handle BC-specific development scenarios. It is a developer-facing tool; its relevance to business users is as a signal that Microsoft is investing in quality-bar infrastructure for AI-generated BC code.
What is MCP and why does BC 28 require it?
Model Context Protocol (MCP) is a standardized protocol that lets AI clients connect to data sources and tools. In Business Central, MCP server access is the connectivity layer that lets external AI tools (Visual Studio Code, Copilot Studio, custom clients) call into BC. In Update 28.0, MCP server access is mandatory and cannot be disabled. The same BC permission model that governs human users governs MCP connections.
When will custom agents be deployable to production?
Update 28.1, which Microsoft MVP Stefano Demiliani has documented is rolling out in May to June 2026, adds production deployment for custom agents built with Agent Designer. Before 28.1, Agent Designer is a prototyping environment. After 28.1, custom agents can run alongside the built-in Payables and Sales Order agents in production environments.
Our Fact Checking Process
We prioritize accuracy and integrity in our content. Here's how we maintain high standards:
- Expert Review: All articles are reviewed by subject matter experts.
- Source Validation: Information is backed by credible, up-to-date sources.
- Transparency: We clearly cite references and disclose potential conflicts.
Your trust is important. Learn more about our Fact Checking process and editorial policy.
Our Review Board
Our content is carefully reviewed by experienced professionals to ensure accuracy and relevance.
- Qualified Experts: Each article is assessed by specialists with field-specific knowledge.
- Up-to-date Insights: We incorporate the latest research, trends, and standards.
- Commitment to Quality: Reviewers ensure clarity, correctness, and completeness.
Look for the expert-reviewed label to read content you can trust.
