What's New: Troubleshooting Business Central Agents
How to troubleshoot Business Central agents with the agent task log and the view details page. It explains how an agent makes decisions (instructions, tools, virtual user interface, task messages, step history). A sales order risk classification agent asks for help because it cannot find a risk field, and the view details sections show that the field is missing and a reason field is not editable, likely because of a profile or per-tenant extension mistake.
- Agent decision-making fundamentals
- Agent example and problem scenario
- Accessing agent logs and task details
- Analyzing agent perception and configuration
- Examining tools, memory, and messages
- Page stack and settings information
- Summary and troubleshooting recap
Overview
The video explains that an agent makes decisions from its instructions, the tools available to it, a virtual user interface representation of Business Central, task messages and step history. It then uses an agent example in which the agent stops and requests assistance because there is a risk field it is supposed to fill.
The demo goes from the agents tasks page to the task log and into view details. There the user can see what the agent saw, which tools it had, what data it memorized, which messages it could access, the page stack, and settings such as currency and message formatting. These help work out why an agent made a given decision and how to adjust its instructions.
Key points
- An agent’s decision depends on five inputs: instructions, available tools, the virtual user interface representation, task messages and step history.
- The agents tasks page gives access to the task log, which lists the steps the agent performed, including assistance requests with a message explaining why help is needed.
- In the example, the agent could not proceed because of a risk field it was supposed to fill, and the log made this visible.
- View details has a section showing a textual view of what the agent saw: page type, editability, description, actions and fields it can interact with.
- Searching that view showed no risk field and a reason field that was not editable. The likely cause was a mistake in the profile or a field never added through the per-tenant extension.
- View details also shows the tools the agent had at a specific point, which helps when designing instructions around available capabilities.
- The memorized data and messages sections show what the agent retained and what messages it could see up to that step.
Features
Every feature below: status not stated, demoed.
| Feature | At |
|---|---|
| Agent decision-making framework | 0:17 |
| Agent task logging and view log entries | 2:16 |
| Agent view details page | 2:45 |
| What the agent saw section | 3:10 |
| What tools the agent had access to | 4:28 |
| What data the agent memorized section | 4:52 |
| What messages the agent had access to section | 5:18 |
| Page stack fact box | 5:39 |
| Agent settings and formatting | 5:52 |
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.
Not found in BC28-30: page “agents tasks page”, page “sales order page”, page “sales order card”.
Quotes
- 0:17 “an agent is a system that takes a few different elements into account to make a decision”
- 0:28 “The first of them is the instructions which is the blueprint how the agent works and what its purpose is”
- 2:02 “the agent is telling me it can’t proceed with the task because there is a risk field that is it is supposed to fill”
- 2:45 “If I click on view details uh I get uh a lot of insight into what exactly went into the decision of the agent”
- 3:10 “I see a textual view of what the agent sees as well. So, uh I can understand exactly what things it can take into”
- 4:05 “I can see that I can find the reason field and uh I can see it’s also not editable”
- 4:18 “Most likely I have made a mistake on my profile or I’ve forgotten to add it to add it via my per tenant extension”
- 4:38 “This section uh gives you the ability to see what capabilities the agent could use at a specific point”
- 6:03 “we can answer what was on a page, what the properties of the elements were on this page”