Agent Tuning Labs

Overview

Agent Tuning labs give you a pre-built AI agent and a scenario to solve. Your job is to adjust the agent's integrations, skills, and model selection to hit the scenario's objectives while staying within a cost cap. As you make changes, you'll see the token cost update in real time  so you can find the right balance between capability and cost.

Before you start

• Agent Tuning is available on Immersive One Advanced.
• No setup or configuration is needed; each lab gives you a scenario and a pre-built agent to work from.
• You can find Agent Tuning scenarios from the Exercise – Agent Tuning navigation item. Managers can also assign Agent Tuning collections to their teams.

What you'll get out of it

• A completion result for each attempt, including your score, an AI summary, and the objectives you reached.
• Your token use and token cost for each attempt, so you can see the financial impact of your choices.
• A view of your best attempt and your progression over time.

How to run an Agent Tuning lab

Watch a quick walkthrough of Agent Tuning Labs: 

This video shows you how to pick a scenario, write your agent’s instructions and skills, choose the right model and tools, and submit a try for grading. It also covers how to read your score report and improve with each attempt.

1. Find an Agent Tuning scenario from the Exercise – Agent Tuning navigation item, or open a collection your manager has assigned.

2. Open the lab detail page to see the learning outcomes, brief, and objectives.

3. Start the lab. On first use, follow the short tutorial that walks you through the screen.

4. Tune the agent: adjust its integrations, skills, and model selection to meet the objectives while keeping within the cost cap. Watch the live cost as you make changes.

5. Submit when you're ready. Adding notes at submission is optional.

6. Review your completion report – it shows your score, AI summary, and which objectives you reached.

7. Try again to improve. Compare attempts to see your best result and track your progress.

What you'll see on screen

The key screens you'll work with, and what to look for on each.

Lab detail page. Before you start, this shows the lab's learning outcomes, a brief, and its objectives. From the "Results" tab you can also see a history of your previous attempts: your best score, token use, token cost, and your progression over time.

First-time tutorial. On your first entry, a short tutorial spotlights each part of the screen in turn. If you're ever unsure what something does, run through it again.

Workspace. This is where you do the work. You adjust the agent's integrations, skills, and model selection to meet the scenario's objectives.

The workspace has the following tabs:

• Briefing – Recap on the challenge, the tools available to you, what good looks like, and what to look out for.
• Chat – Interact with and prompt the agent to carry out its defined tasks.
• Previous Tries – View a history of your previous tries within the current session.
• Report – See a real-time view of token cost, LLM calls, and tool calls as you work. Once you've submitted, come here to see your score and performance summary.
• Live cost. As you change the model or tune the agent, the cost updates in real time in the top right of your workspace so you can see the financial impact of your choices.

Completion page. After you submit, you'll see your score, an AI summary, and the objectives you reached, alongside your token use and cost. Use it to decide what to change on your next attempt.

Troubleshooting