Adding one missing objection to a roleplay agent used to mean finding the right persona field and editing it by hand. Now it means typing "add an objection that there's no budget this quarter" into a box next to the agent and hitting submit.
Why this matters
An agent can start from a prompt, a template, a real call, or an uploaded file, and each of those methods produces a working persona fast. The friction shows up afterward, the first time a manager watches a practice session and notices something specific is missing or slightly wrong: the buyer never raises a budget objection, the tone reads as too agreeable, a technical detail is off.
The standard fix was to open the underlying agent configuration and hunt through separate fields for objections, tone, and conversation style, then edit each one manually. That is manageable for a first build. It is a real drag on the tenth small adjustment, especially for an enablement manager without the time or inclination to learn where every setting lives.
How it works
After an agent is generated, by any method, Outdoo shows a "Fine tune your AI Agent" panel next to the agent preview. It contains a plain-text box labeled "Add instructions below to refine your agent," with an example placeholder like the budget objection above.
A manager types an instruction in ordinary language, submits it, and the agent updates accordingly, no need to locate the specific field that governs objections or rebuild the persona from scratch. A "Try Out Agent" button sits right next to the panel, so the manager can start a test roleplay immediately and confirm the change actually landed the way they meant it, in the same session rather than after a separate save-and-reload step.
An "Add Context" option handles the case where the fix is less an instruction and more a document, feeding in extra grounding material without rewriting the whole original prompt. Each fine-tune pass is saved as its own version, visible through a version selector such as "Version 3," so a manager can see what the agent looked like before the latest change and compare it against what it looks like now.
What this unlocks
A manager running a live team practice session who notices mid-review that the agent glosses over procurement pushback can fix it on the spot, type the instruction, hit Try Out Agent, and have the corrected version ready before the next rep takes their turn, instead of filing the fix for later.
An enablement lead maintaining a library of dozens of agents across different courses can push small corrections directly, in plain language, without routing every tweak through whoever originally built the agent or understands its field structure.
And because every instruction creates a new version rather than overwriting the last one, a manager who tightens an agent's compliance language for a regulated product can compare the version before the change against the version after it, confirming the edit did what it was supposed to before rolling it out to the whole team, and rolling back to the prior version if it did not.
See Creating roleplay agents: overview for how agents are built before they reach this stage.
The fastest way to correct an agent that is almost right is rarely to rebuild it. It is to tell it, in the same sentence you would use with a colleague, what is still missing.
See it in Action
Tell us what you're working on, whether that's sales calls, support tickets, or leadership coaching, and we'll show you how it fits.
Talk to Sales