topic hubCopilot & Agentsofficial - Microsoftreviewed - checked by Opus

Designing and coding agents (preview)

Designing and coding agents (preview) covers how to design, create, configure, run, iterate and code custom Business Central agents with the AI development toolkit in sandboxes. It answers questions on instructions, permissions and profiles, tasks, attachments, visibility, export/import and the Tasks AL API.

Path: Integration with AI > Designing and coding agents (preview) · tier official · system copilot · narrative reviewed (checked by Opus)

Overview

This section explains how partners prototype and build AI agents in Business Central using the AI development toolkit. The toolkit is sandbox-only and follows an agent lifecycle: create, configure, activate, iterate, graduate and clean up. A transparency note describes the intended use and the Azure OpenAI models behind it.

The pages follow that lifecycle. Start with the overview, then “Create and activate an agent” and “Write effective instructions”, with “Best practices for designing an agent” and “Instruction keywords” as references. Next come permissions and profiles, running tasks, attachment limits, and iterating with the Agent Task Log and instruction version history. The Sales Validation Agent is a preconfigured example.

For developers, “Integrate with the Tasks AL API” shows how to detect agent sessions and read agent information. The subtopic “Coding agents in AL” covers building and registering agents in code, including converting a prototyped agent. Export and import moves configurations between sandboxes, and the FAQ covers common questions.

Key points

  • Agents are prototyped in sandbox environments only, using the AI development toolkit (preview, 27.2 for create and activate).
  • Effective agent permissions are the intersection of the user’s and the agent’s permissions, so an agent never exceeds the scheduling user’s privileges.
  • Profiles and page customization properties such as ClearActions and ClearLayout control what the agent sees and can do in the UI.
  • Tasks are the unit of work; trigger them manually from the Agent Tasks page or through the Tasks AL API from UI actions and business events.
  • Attachments support PDF, PNG and JPG, with file size limits and a 10-page PDF limit; custom processing can use Azure AI Document Intelligence.
  • Iterate using the Agent Task Log, the page accessibility view of what the agent perceives, and instruction version history with comparison and rollback.
  • Agent configurations (identity, instructions, permissions, profile) can be exported and imported as XML between sandboxes.
  • Agent visibility in the role center depends on app installation, Copilot capability status, permissions and archiving state (28.1).

Subtopics

More Learn pages

Videos, posts and code changes

Linked by a Haiku matcher with a grounding quote from the item’s summary (link/topics.ts); machine-generated.

Source: Microsoft Learn (CC BY 4.0). Descriptions are Learn’s own.

What Learn documents here 8 of 23 Learn pages

Show 15 more