Introducing: Envision, Design and Code Business Central Agents
Business Central agents: how the Sales Order Agent and Payables Agent work, the agent framework with human-in-the-loop review, security and admin controls, and two ways to build agents. The Agent Designer in the web client is in preview and works in sandboxes. Agents can also be created in code with the AI development toolkit.
- Introduction to Agentic AI and Current Agents
- Agent Framework and Human-in-the-Loop Experience
- Agent Navigation, Security and Administrative Controls
- Agent Extensibility and Partner Opportunities
- Designing Agents in the Web Client
- AI Development Toolkit and Code-Based Agent Creation
- Agent Use Cases and Scenarios
- Adoption Paths and Getting Started
Overview
The video covers the agentic AI approach in Business Central. It starts with the two existing agents, the Sales Order Agent and the Payables Agent, and then explains the framework behind them. Users see agent tasks in a prominent pane with a timeline, reasoning tooltips and processed documents. Agents navigate the UI like users do.
From an administration standpoint, agents are treated like users, with permissions and profile assignments. They are distributed as apps and enabled or disabled the same way as other apps. The video then shows two ways to build agents. The Agent Designer in the web client is in preview and lets consultants, product owners, domain specialists and makers design proof-of-concept agents in a sandbox. The AI development toolkit lets developers create agents in code. It closes with use cases and adoption paths, including upcoming evaluation tooling and calling BC agents from Copilot Studio via MCP.
Key points
- The Sales Order Agent processes requests for quotes and creates sales orders. The Payables Agent processes vendor invoices and matches them to purchase orders. The Sales Order Agent was shown on a slide and the Payables Agent was demoed in the product.
- Human-in-the-loop: new agent tasks appear in a prominent pane. Users can see the timeline of agent activity, reasoning tooltips and processed documents, and can navigate to verify decisions.
- Agents are built like users for administration. You assign a profile and limit access with permissions, and explicit access controls define which tables and UI elements they can reach.
- Agents are distributed as apps and enabled or disabled with the same mechanism administrators already know.
- The Agent Designer in the web client is in preview. It is a wizard supported in sandboxes, for proof-of-concept agents with instructions, profile, permissions and manual trigger simulation.
- In the preview, agents created in code with the AI development toolkit deploy to sandboxes only. Production deployment is stated as coming later.
- Evaluation of agent results is stated as coming soon. Calling BC agents via MCP from Copilot Studio is a future capability.
Features
| Feature | Status | At | Evidence |
|---|---|---|---|
| Sales Order Agent | status not stated, demoed | 0:47 | |
| Payables Agent | status not stated, demoed | 1:03 | |
| Human-in-the-Loop Agent Experience | status not stated, demoed | 2:24 | |
| Agent UI Navigation Framework | status not stated, demoed | 4:06 | |
| Agent Security and Access Control | status not stated, demoed | 5:49 | |
| Agent App Distribution | status not stated, demoed | 6:54 | |
| Agent Designer in Web Client | preview, demoed | 9:21 | “we are now launching in preview the ability for you especially consultants, product owners, domain specialists and and even makers to design proof of” (9:21) |
| AI Development Toolkit - Agent Code Creation | status not stated, demoed | 12:23 | |
| Agent Extension Deployment | preview | 13:14 | “In the preview, we support that you can deploy to sandboxes. Uh later you will of course be able to deploy uh these extensions” (13:14) |
| Agent Evaluation Framework | status not stated | 13:42 | |
| Copilot Studio Agent Integration | announced | 17:06 | “in the future even call BC agents uh via MCP from within copilot studio” (17:39) |
AL objects mentioned
As heard in the captions. A name that matches one object page by exact type and name links to it; the others stay as named.
Quotes
- 0:47 “For a while now, two fantastic agents are helping two of your teams, the sales team and the payables team to enhance their operations.”
- 9:21 “we are now launching in preview the ability for you especially consultants, product owners, domain specialists and and even makers to design proof of”
- 9:50 “you can assign a profile to the agent. You can limit its access by giving it permissions.”
- 13:14 “In the preview, we support that you can deploy to sandboxes. Uh later you will of course be able to deploy uh these extensions”
- 14:10 “scenarios where you need to handle unstructured inputs. That’s an obvious candidate. Um AI and LLMs are great at natural language, emails, documents, etc.”
- 15:02 “Agents work really well when they can act as a user. Uh having the permissions like a user, restricted access to to data, etc.”
- 17:06 “you have a choice of both native BC agents as well as uh no and low code copilot studio agents and this is really”
Disclaimers in the video
- 9:21 preview: we are now launching in preview the ability for you especially consultants, product owners, domain specialists and and even makers to design proof of concept business central agents
- 10:29 preview: this experience is an incline experience and it’s supported in sandboxes
- 13:14 coming-later: In the preview, we support that you can deploy to sandboxes. Uh later you will of course be able to deploy uh these extensions to production environments
- 13:42 coming-later: we will also soon give you the means to actually go and evaluate the agents so that you can automate evaluating the results
- 17:39 coming-later: in the future even call BC agents uh via MCP from within copilot studio
Presenters (as heard): Dmitri Richard, Peter Boring.