Most of the work people hand to Xagent starts as one sentence. Ours, for this walkthrough: “Search the most important AI news from the past week and turn it into a short two-host podcast.”
That sentence is the whole brief. You don’t draw a flow or pick tools first. You type it into the task box, choose a model if you want a different one from the default, and send it.
The plan comes first
Before anything runs, Xagent writes a plan. On the right of the task screen you see it as an execution graph, one card per step. For the podcast it came out as three steps: research the week’s news, write the dialogue script, then turn the script into audio.
Each card shows the tool that step will use and its status. You can switch the graph between a vertical and a horizontal view. The vertical view is easier to read on a laptop screen.
Every tool call is on the screen
When the run starts, the conversation in the middle fills up with tool calls. For this task that meant a web search that ranked 18 stories and kept 5, a Python step that drafted a four-minute script for two hosts, and a text-to-speech step that produced the audio file.
Nothing happens out of sight. Every call shows what went in and what came back, and the token count for the run sits under the task box. If something looks off halfway through, you can pause the task and look at it before it goes further.
The result, and the trace behind it
The podcast came back as an MP3, four minutes and eleven seconds long. That’s the part people notice. The part admins care about is the trace: every step, every tool call and every result for the run, kept so you can go back and check how the output was made.
Keep what worked
If you’re going to want the same podcast next Friday, save the task as an agent. The agent keeps the role, the model, the tools and any knowledge you attached. Next week you run it again, or put it on a schedule so it runs itself.
Start with one task you already do every week. That’s usually the easiest one to judge, because you know what a good result looks like.
