← Business Central Approvals collection
Business Central AI Agents Agent Governance Decision Infrastructure

Business Central AI Agents Don't Answer to the Same Rules. You Do.

Payables Agent, Copilot, Agent Designer, Power Automate, and OpsGrid each write to, or route decisions around, Business Central now. Only one of them was built to say who approved it, the same way, every time.

Christopher Wakare
8 min read
Business Central Approvals

A $34,000 vendor invoice posts to Business Central on a Tuesday morning. By Thursday, finance still can't say who approved it. The Payables Agent's review screen shows a "reviewed" action logged by someone who left the company in June. A Power Automate flow that used to route vendor approvals into Teams says it never touched this vendor. Nobody thinks to check whether the invoice actually came through a custom agent someone built in Agent Designer the week before, because nobody remembers Agent Designer exists until a colleague mentions it in the meeting called to sort this out.

The invoice itself was fine. The three days spent reconstructing who approved it were not. That's the cost of running five separate AI and automation surfaces against one Business Central tenant, as of August 26, 2026, with no shared answer to one question: which tool actually touched this record, and who's the named person accountable for it.

Business Central agent governance, separate from what any single tool's settings screen shows you, is the shared answer to three questions across every AI surface writing to your ERP: which system made the write, who's the named person accountable for it, and where the record lives. Business Central runs five of these surfaces today: Payables Agent, Dynamics 365 Copilot, Agent Designer, Power Automate, and IntelliConnectQ's own OpsGrid. Only one was built to answer all three questions in the same place, every time.

Disclosure: IntelliConnectQ Analytics builds OpsGrid, one of the five tools compared below. This field guide is written to be accurate and checkable, not promotional. Every date and capability claim about Microsoft's tools links to Microsoft's own release-plan documentation, not a summary of it.

The five AI and automation tools now touching Business Central

Five surfaces write to, or route decisions around, a Business Central tenant today. Two are built-in agents. One is a conversational assistant. One is a do-it-yourself automation platform. One is a governed decision layer purpose-built for BC. Here's what each one is, and how mature it actually is, not how mature the announcement made it sound.

Tool 1
Payables AgentBuilt-in agent that reads incoming vendor invoice emails, extracts the data, and matches it to purchase orders, preparing a draft for review. Reached general availability November 7, 2025; its purchase-invoice matching capability specifically reached GA February 6, 2026.
Tool 2
Dynamics 365 CopilotGeneral-purpose conversational assistant across Business Central and the rest of Microsoft 365. You ask, it answers or drafts. Licensed per user, per month, on top of an existing M365 subscription. Long-standing, generally available product.
Tool 3
Agent DesignerMicrosoft's in-product, natural-language builder for custom Business Central agents, sandboxed to BC data. Enabled for public preview February 6, 2026, with general availability listed for October 2026.
Tool 4
Power AutomateMicrosoft's own suggested workaround for the gaps native BC workflows don't cover, Teams approval cards included. Whoever builds the flow owns its trigger logic, its routing rules, and, if they leave, its documentation, or the lack of it, as IntelliConnectQ has written about before.
Tool 5
OpsGridIntelliConnectQ's own governed decision layer for Business Central, in active beta. Monitors BC continuously, ranks signals by cost impact, and routes them to a named owner in Microsoft Teams. Nothing posts back to BC without that named owner's explicit approval.

Where each Business Central AI agent's authority actually stops

Five tools, five different answers to who's allowed to do what, and who has to answer for it afterward. The table below lines them up across the dimensions that matter once something goes wrong: not what a tool does when it works, but what it produces when someone has to check.

Dimension Payables Agent Copilot Agent Designer Power Automate OpsGrid
Purpose Extracts invoice data, matches to POs, drafts for review Conversational Q&A and drafting across BC data In-product builder for custom BC agents DIY flows scripting routing around BC events Governed decision routing across BC and systems BC can't reach
Interaction model Automatic trigger on incoming invoice email User-initiated, reactive chat Manual invocation only, no auto triggers yet Depends on the flow's own trigger logic Proactive monitoring, routes ranked signals to a named owner
Human-in-the-loop Draft reviewed before posting, per Microsoft's agent runtime User reviews before acting; no autonomous write path Whatever the builder configures, no enforced default Whatever the builder configures, not platform-enforced Architectural requirement: no BC write without named-owner approval
Governance / audit trail BC change log plus agent timeline (execution log, not a named-approver record) No dedicated audit record beyond the standard BC transaction log Diagnostic and troubleshooting logs, built for developers Flow run history, rarely reviewed unless something breaks Signal, approver, action, timestamp: one record, by design
Pricing / billing model Copilot credits, prepaid or pay-as-you-go Per-user, per-month M365 Copilot license Copilot credits, same model as built-in agents Premium connector cost, scales per user Active beta, pricing not published
Maturity status GA, Nov 7, 2025 (PO matching GA Feb 6, 2026) GA, established product Public preview since Feb 6, 2026; GA planned Oct 2026 GA, established product Active beta, not GA
Feb 6
2026: Agent Designer's public preview started
Oct
2026: planned GA, per Microsoft's release plan, which flags that delivery timelines may change
0
Built-in triggers for custom agents today. Manual invocation only, or developer-wired via the AI Development Toolkit APIs

The gap in that Maturity row is the one worth sitting with. Two of these five tools have been generally available for less than a year. A third is still in public preview with no committed date, and Microsoft's own release plan won't commit to October holding. Building a governance model today around all five assumes a stability none of them have actually earned yet, except for Copilot and Power Automate, old enough that most BC teams have already made peace with their limits.

Business Central agent governance doesn't exist as a shared model

Here's the assumption worth challenging: that adding more AI coverage inside Business Central closes gaps. It closes capability gaps. It opens a coordination gap, because none of these five tools were built to agree with each other. Microsoft built a real write-scope control, the MCP Server Configuration, and a human-in-the-loop approval requirement inside the agent runtime. Each of those controls operates one tool at a time. Nothing forces the Payables Agent's execution log, a Power Automate flow's run history, and a custom Agent Designer agent's diagnostic trace into one format a compliance team, or an auditor, can read without translating three of them first.

OpsGrid answers those three governance questions the same way every time, by design. A Signal is detected in BC, Routed to one named owner, held for Approval before anything writes back, Executed only after that approval, and Audited as a single record naming who approved what, and when. That sequence doesn't make OpsGrid better at reading an invoice than the Payables Agent, or better at answering a question than Copilot. It makes OpsGrid the one tool among the five built to answer the accountability question the same way on every record, not only the ones someone remembers to check.

Free · 3 minutes
Which of your five BC tools is actually the accountability gap?

The Decision Latency Diagnostic scores your Signal, Route, Approve, Execute, and Audit stages in 12 questions, across whichever of these tools are touching your BC data today.

Take the diagnostic →

Which of the five fits the job you actually have

Native and Microsoft tools cover you when
One approval chain, low change frequency, and an IT team, not one analyst, owns whatever flow gets built
Your gap is findability: the data's already in BC, someone just has to ask the right question
You need invoice or sales-order drafts prepared automatically and reviewed by the person who already reviews them
Add a governed layer when
Decisions span BC and systems BC can't reach: freight portals, WMS, supplier systems, Teams approvals
Spend or compliance exposure needs one named-owner audit record, not three tools' worth of logs to reconcile
You've already lived a version of the invoice story above, and don't want a second one

Frequently asked questions

Is Business Central's Agent Designer generally available yet?
No. Microsoft enabled Agent Designer for public preview on February 6, 2026, and its own release plan lists general availability for October 2026, not before. Microsoft's release plan carries a standing disclaimer that delivery timelines may change and projected functionality may not be released, so treat that October date as a target, not a commitment. As of August 26, 2026, any Business Central agent built in Agent Designer is a preview-stage tool.
Do Business Central's built-in AI agents require human approval before they act?
Microsoft's own agent runtime documentation says every Business Central agent, built-in or custom, operates through human-in-the-loop approvals for critical operations, with actions logged and permissions enforced. What counts as critical, and what the approval screen actually shows the approver, is set independently inside each agent's own configuration. Nothing forces the Payables Agent, the Sales Order Agent, and a custom Agent Designer agent to define critical the same way, or to log the approval in a format anyone outside that one tool can read.
What's the difference between Business Central's Agent Designer and Microsoft Copilot Studio?
They're different layers, by Microsoft's own description. Business Central agents, including custom ones built in Agent Designer, run inside the product on Business Central's own security model, permissions, and data. Copilot Studio is a separate, cross-product orchestrator: Microsoft's release plan says Business Central agent capabilities will, in the future, be exposed as MCP tools for Declarative Agents built in Copilot Studio to consume across Microsoft 365. Today, an Agent Designer agent is BC-native. A Copilot Studio agent orchestrates across products, including BC, from outside it.
Is OpsGrid generally available for Business Central?
No. OpsGrid is in active beta with a focused group of mid-market manufacturers, distributors, and logistics operators running Business Central. Pricing has not been published. Beta access starts with a free Decision Latency Audit at intelliconnectq.com/opsgrid.

Five tools now touch Business Central's data. Only one of them was built to answer the same accountability question about every single write. That's not a knock on Microsoft's agents; each does its specific job well. It's the reason "we added AI to Business Central" and "we can name, in one place, who approved that write" are still two separate projects, at least as of August 26, 2026.

Not sure which of your five BC tools needs the governance layer?

Take the Decision Latency Diagnostic

12 questions, 3 minutes. See where your Business Central signal-to-action gap actually sits, across whichever of these five tools you're already running.

Start the diagnostic →

The Execution Edge

Monthly. For operations leaders building faster on AI. Real case studies, system blueprints, and tools — no fluff.

Your subscription could not be saved. Please try again.
Your subscription has been successful.