Guide
Roleplay Agent Creation 101: The Field Guide
Building an AI roleplay is easy. Building one that gives your team real practice, and coaching that changes the next real call, is the harder part.
This is the Roleplay Agent Creation 101 workshop written down: the six layers of a roleplay agent in build order, the three places most of them break, a worked scorecard for an enterprise discovery call, an honest comparison of four ways to build the same agent, and a prompt pack of eight prompts you can run tomorrow.
Key Takeaways
- The six layers of a roleplay agent in build order - persona, scenario, objections, resources, scorecard, coaching — and why most agents fail on layers 3, 4 and 5, not on the persona
- How the same resource library grounds both the buyer and the rubric, which is what makes a score mean something
- A worked scorecard for an enterprise discovery call: eight weighted criteria, what "good" looks like for each, and the document behind it
- When to use Standard, Strict and Stringent scoring, and how to sanity-check your weights by scoring a deliberately bad run
- A side-by-side look at four ways to build the same agent - ElevenLabs, Claude Code, ChatGPT and Outdoo, and where each one stops
- The prompt pack: four master prompts (agent spec, scorecard, critic, variations) plus one adaptation each for ElevenLabs, ChatGPT, Claude Code and Outdoo
