Putting Business Central agents to work, with people in charge

What Business Central agents are and how they work

An agent is set up inside Business Central with its own identity, its own permissions and written instructions. It watches a trigger, typically a shared mailbox. When something arrives, it works through the same pages a user would. It finds the customer, checks item availability and fills in a document.

As of October 2026, Microsoft documents three agents: the Sales Order Agent, the Payables Agent and the Expense Agent. Which ones you can use depends on your country version and language, and Microsoft's availability page on Learn lists the current status. Each step an agent takes is recorded on a timeline. It also stops to ask for help or approval at set points, instead of guessing.

Which process can Business Central agents handle well?

Agents do best where requests are frequent, similar and low in risk. A distributor that gets many emailed orders for catalog items is a natural fit. A company that sells engineered products with negotiated terms on every deal is not. Each of those requests needs judgment that an agent does not have.

So we look at a sample of real incoming messages with your team. Then we sort them. Some an agent could complete. Some it could start and hand over. Others it should leave alone. The sample also shows how customers and vendors actually write. That shapes the instructions we give the agent.

Preparing data before Business Central agents touch it

An agent can only match what it can find. Suppose a customer emails from an address that sits on no contact. Or suppose the message asks for an item by a name that appears nowhere on the item card. Then the agent stalls. So contact records, item descriptions, item references and units of measure need attention first.

Pricing and availability settings matter too. When price lists, discounts and order promising are set up properly, the quote an agent prepares matches what a person would offer. If they are not, the agent faithfully repeats the gaps. We fix those foundations as part of the setup.

Permissions and approval points for Business Central agents

An agent should hold the narrowest permissions that let it finish its task. One that drafts sales quotes has no reason to read payroll dimensions or post journals. So we build a dedicated permission set. Then we test it as the agent and confirm what is out of reach.

We also set approval points. Anything that leaves the company, such as a quote sent to a customer, is seen by a named person first. That person can correct the draft, and the correction stays visible afterward. We check outgoing messages for tone and accuracy, since the agent writes in your name.

A pilot that shows whether the agent earns its place

We start in a sandbox with copies of real messages. Then we move to production for one mailbox or one customer group. During the pilot, supervisors note every case where they had to step in and why. Patterns in those notes lead to clearer instructions or better master data.

Microsoft bills agent activity on consumption, and the current terms are on its site. So we track usage next to results. At the end of the pilot, you have evidence. You can see how many requests the agent handled and how many needed correction. You also learn whether staff want to keep it.

Business Central agents: what companies ask

How are Business Central agents different from Copilot?

Copilot features help a user with one step when asked. Examples are drafting a product description or suggesting bank reconciliation matches. An agent runs a whole task by itself across several pages after a trigger. It then brings a person in at approval points. So Copilot supports your staff, while an agent takes over a repeat job and asks for a check.

Can an agent post transactions without anyone checking?

Agents are designed to keep people involved. They prepare documents and pause for confirmation at sensitive steps. Their permissions also decide what they can do at all. We set up both, so that nothing reaches a customer, a vendor or the ledger unseen. You can then see, step by step, what the agent did and who approved it.

Do Business Central agents work on-premises?

Microsoft delivers its agents and Copilot features through Business Central online. A company on-premises that wants them generally needs to move to the cloud first. We can assess that move for you. We can also tell you what else would change along the way, such as extensions, integrations and reports.

Can we build our own agent for a process Microsoft does not cover?

Microsoft's platform allows custom agents that work with Business Central data through its APIs. Before we build one, we check whether a workflow, a Power Automate flow or a small extension would solve the problem more simply. An agent adds running charges and oversight work. So it should be the choice only when simpler tools fall short.

What are the risks of using agents?

An agent can misread an ambiguous message, match the wrong record or word a reply badly. The safeguards are narrow permissions, approval before anything goes out, a visible activity log and a pilot period. With those in place, mistakes are caught at the draft stage. Staff then correct them, and the notes feed back into the instructions.

Related automation services

Business Central agents build on other automation in Business Central. These pages cover Copilot features, Power Automate flows, sales order processing and accounts payable automation. Those are the areas where an agent is most likely to help. If you want to test one process, the contact page is the place to start the talk. For the product detail, see Microsoft Learn: Copilot and agents availability in Business Central.