Use the AI Agent Task to let Assistant “figure out” a small job from a plain English description. You tell it what you want; it can use available tools (if you allow them) to help get it done.
This feature is experimental. Start simple, see what it does, then refine your wording.
IMAGE_PLACEHOLDER: overview-screenshot.png
Caption: Overall AI Agent Task screen with Instructions, MCP Tools, and Available Servers panels highlighted.
Not ideal for:
IMAGE_PLACEHOLDER: use-cases-callout.png
Caption: A simple callout panel listing common good uses vs when not to use.
IMAGE_PLACEHOLDER: create-task.png
Caption: Creating a new AI Agent task in the task list.
IMAGE_PLACEHOLDER: instructions-box.png
Caption: The Instructions box with an example prompt entered.
Be clear, specific, and include any limits or desired format. Examples:
Helpful tips:
IMAGE_PLACEHOLDER: good-vs-vague-instructions.png
Caption: Comparison of a vague vs an improved instruction.
Tools = capabilities (like reading files, syncing, model actions). The agent can only use tools you include.
In the AI Agent Task screen you’ll see:
How to add tools:
Search bar:
You can still run the task with no tools selected—the agent will try to answer using only general AI knowledge, but it obviously can’t read files or perform system actions without the right tools enabled.
IMAGE_PLACEHOLDER: add-built-in-tools.png
Caption: Adding built-in tools via the Available Servers panel.
IMAGE_PLACEHOLDER: server-expanded-tools.png
Caption: A server expanded showing a list of selectable tools.
IMAGE_PLACEHOLDER: search-tools.png
Caption: Using the search box to filter tools by name.
The output is a conversation-style log so you understand what happened. Typical lines:
user: Your request.Running Tool: <name> The agent decided to use a tool.tool: Received response The tool replied (sometimes simplified).assistant: The agent’s messages back to you.Final lines usually contain the actual answer.
If it didn’t use a tool you expected:
IMAGE_PLACEHOLDER: result-log.png
Caption: Example result log showing tool calls and assistant responses.
| Goal | Instructions |
|---|---|
| List files | “List all files in the Models folder and show size (KB). Return as a table.” |
| Compare sources | “Compare filenames in ACC ‘/Models’ and SharePoint ‘/Models’. List ones missing from SharePoint.” |
| Simple script | “Generate a Python script that prints room numbers 101–150.” |
| Check software | “Check if Revit 2024 is installed. Reply only: Installed or Not Installed.” |
| Refine output | “Summarize the tools you used. Return a bullet list.” |
IMAGE_PLACEHOLDER: examples-collage.png
Caption: Collage of different instruction examples in the UI.
Common refinements:
Run, adjust, repeat. That’s expected.
IMAGE_PLACEHOLDER: refinement-iteration.png
Caption: Iterating on a prompt from vague to precise.
| Issue | What to Do |
|---|---|
| “Instructions cannot be empty.” | Enter something—must not be blank. |
| It ignored a tool | Mention it explicitly or ensure it’s selected. |
| Tool you expect isn’t listed | Add or configure that server first. |
| Output too generic | Ask for specifics (format, fields, limits). |
| No tool used when needed | Say: “Use the X tool to …” |
| Nothing happens for a long time | Reduce number of selected tools or simplify request. |
| Strange answer | Refine wording; remove ambiguity. |
IMAGE_PLACEHOLDER: troubleshooting-panel.png
Caption: Troubleshooting tips sidebar mockup.
IMAGE_PLACEHOLDER: good-habits.png
Caption: Quick habits reminder graphic.
IMAGE_PLACEHOLDER: safety-callout.png
Caption: Safety and privacy callout banner.
When a prompt + tool combo works reliably:
IMAGE_PLACEHOLDER: convert-to-standard-task.png
Caption: Converting a refined agent task into a standard task.
IMAGE_PLACEHOLDER: checklist.png
Caption: Visual checklist overlay.
This feature is evolving. Let us know:
Your feedback helps improve future versions.
IMAGE_PLACEHOLDER: feedback-button.png
Caption: Mockup of a feedback button or dialog trigger.