How to build a coaching agent in Proshort
Build a coaching agent that scores reps' calls on a skill you choose: instructions, golden pitch, whose calls it scores, a scoring rubric and role-play.
Watch on YouTube →Part 1 of the coaching agents guide. Coaching agents are how Aria scores reps' calls on the skills you care about. If your team's My skills page is empty, usually no agent has been set up yet. Only admins can build one.
What the video covers
Admin console > AI Agents > Create agent: answer three questions and click Generate agent.
Name the agent, edit the instructions, and set a golden pitch, written or from a real recording.
Choose whose calls it scores (a team such as BDR), when it runs (deal, contact or lead calls) and what context it may read; test it on a real call.
Generate a five-band scoring rubric, turn on role-play with a persona and passing score, then Publish.
Chapters
0:00 Open AI Agents
0:23 Your agents
0:33 Answer three questions
0:50 Name and instructions
1:04 Golden pitch
1:18 Whose calls it scores
1:29 When it runs
1:45 Context and test
1:59 Scoring rubric
2:22 Role-play
2:36 Publish
Transcript
Hi, I'm Aria. Coaching agents are how I score your reps' calls on the skills you care about. If your team's My skills page looked empty, this is usually why: no agent was set up yet. Only admins can do it. Open the admin console and click AI Agents.
Every agent you've published is listed here with its type, status, how many calls it has run on, and when it last ran. Click Create agent to add a new one.
I'll ask three questions. Pick Coaching agent, type the skill you want scored, like budget qualification, and tell me what good looks like, or leave it to me, then click Generate agent.
I draft the agent for you. Give it a clear name, because reps will see it on their scorecard. The instructions describe the behaviour I detect on a call. Edit them, or click Rewrite with AI to sharpen the wording.
The golden pitch is the benchmark for a five out of five call. Write it here, or attach a real recording from your library and I'll learn from it. Click Continue to set who this agent scores.
Step two decides whose calls I score. Choose a team, such as BDR, and only their calls run through this agent. Teams come from each person's department in Manage users.
Then pick when it runs: calls linked to a deal, a contact, or a lead. For a BDR team that's usually leads. Add filters if you want to narrow further, for example a specific group or region.
Context is what I'm allowed to read: the meeting transcript, the deal's emails, and CRM fields. Before you move on, use Test on a real call to preview what this agent would produce.
Step three is the scoring rubric. Click Generate evaluation and I write five bands, each anchored in observable behaviour, so scores are consistent and defensible. Edit any band, or regenerate the whole rubric from your instructions.
Step four is role-play. Turn it on and reps can practise this exact skill against an AI buyer. Pick a default persona, like the distracted buyer, set the passing score, and add a scenario if you want one.
Click Publish. The agent starts scoring the next qualifying call, and reps see it on their skills page.
