Back to Blog
Published:
Last Updated:
Fresh Content
Agentic ERPFoundational guide

Microsoft's Dynamics 365 agents vs a custom build: what each covers, and the gaps between them

3 min read
642 words
high priority
Ali Ahmed

Ali Ahmed

AI Solutions Engineer, Cognilium AI

TL;DR

Microsoft ships first-party agents for Dynamics 365 — reconciliation, finance, procurement, payables, expense. Here's what each covers, where they stop, and when a custom build is the right call for your processes.

Microsoft is shipping real agents for Dynamics 365, not slideware. If you're deciding where AI goes in your ERP, the first useful step is to know what Microsoft already gives you — because for a lot of standard processes, the answer is "use theirs," and a good partner will tell you so.

Here's the honest map: what the first-party agents cover, where they stop, and when a custom build is the right call.

What Microsoft ships (use these where they fit)

Across Finance, Supply Chain and Business Central, Microsoft's first-party agents already target real, named processes:

  • Account Reconciliation Agent — resolving reconciliation exceptions, increasingly in bulk, enabled by legal entity and module.
  • Finance Agent — skills across reconciliation, data preparation, variance analysis, collections and business intelligence, surfaced through Copilot, Excel, Outlook and Teams.
  • Supply-chain agents — including procurement-change impact analysis (in preview) that traces a supplier's proposed PO change through to affected sales, production, transfer orders and projected inventory.
  • Business Central (mid-market) — a Payables Agent reconciling purchase orders against item amounts and expected goods receipts, and an Expense Agent checking expense reports for policy violations or suspicious entries.

If your process is one of these and runs close to standard, start with Microsoft's agent. It's the fastest, lowest-risk path, and it's the native one.

The right mental model for all of them is operate by exception: the agent clears the routine cases so your team only touches the ones that need judgement. "Reconciliation that clears its own exceptions" is a sharper promise than "AI that automates finance," and it's what these actually do.

Where the first-party agents stop

Three gaps are structural, and no amount of waiting for the next release closes them for you:

  1. Your customizations. First-party agents reason over standard Finance & Operations. The X++ customizations, ISV logic and custom entities that make your estate yours are invisible to them — so they handle the textbook cases and hand back the ones touching your custom process. (The fix is exposing that logic as governed MCP Action tools.)
  2. Your specific processes. Microsoft builds agents for common, named processes. The process that is your competitive edge — your quoting logic, your pricing decision, your bespoke approval chain — is not on anyone's roadmap but yours.
  3. Governance of your estate. An agent inherits the permissions of the user it acts for, and the controls to constrain that are still in flight. The state of your own role design, and scoping agents to only the actions they should take, remain yours to own. (See the Dynamics 365 agent governance gap.)

When a custom build is the right call

Build custom — on top of Microsoft's platform, never replacing it — when:

  • Your F&O is meaningfully customized and the agents keep stalling on the parts that matter.
  • The decision you want computed is specific to you — pricing, demand/inventory, pick-path, quoting — and Microsoft doesn't ship an agent for it (these are AI optimization apps, a different category from the exception-clearing agents).
  • You need governance and audit tighter than the out-of-the-box controls: approval before posting, segregation of duties, every action traced to the instruction and the record.
  • You want to codify your own procedures as reusable, governed context — your playbook, as business skills — rather than rely only on generic behaviour.

The decision in one line

Use Microsoft's agents for standard, named processes; build custom for your customizations, your specific decisions, and the governance of your own estate. It's not either/or — the custom work sits on top of the first-party platform and often makes Microsoft's own agents usable on your customized F&O.

If you're mapping which of your processes Microsoft already covers and which need a build, tell us what you run and where the agents stall — you'll get a straight read, including "Microsoft's agent already does this" when it does.

Share this article

Ali Ahmed

Ali Ahmed

AI Solutions Engineer, Cognilium AI

Ali Ahmed is an AI Solutions Engineer at Cognilium AI.

Applied AI AgentsAgentic SystemsRetrieval-Augmented Generation (RAG)LLM Product Engineering

Frequently Asked Questions

Find answers to common questions about the topics covered in this article.

Among others: an Account Reconciliation Agent and Finance Agent (Finance), procurement/supply-chain agents, and for Business Central a Payables Agent and an Expense Agent. They target standard, named processes.
Use Microsoft's for standard processes. You need custom when your F&O is heavily customized, when the decision is specific to your business (pricing, inventory, quoting), or when you need governance and audit beyond the out-of-the-box controls.
They work on the standard surface; your X++ and ISV customizations are invisible to them. Exposing that logic as governed MCP Action tools is what lets agents work on your customized estate.
No. It sits on top of the platform and often makes the first-party agents usable on your customizations — it complements them, it doesn't replace them.
The agent clears the routine cases automatically so your team only handles the ones that need judgement — e.g., reconciliation that resolves its own standard exceptions and escalates the rest.

Still have questions?

Get in touch with our team for personalized assistance.

Contact Us

Still have a question this did not answer?

The person who wrote this article answers these. Describe your setup and what you are stuck on — you will get a straight answer, including where we think the approach is wrong.

Questions on this?

Talk to an engineer

Schedule a call