Every sales team hits the same wall: the new hire has memorized the playbook and read the product docs, but still freezes in front of a real customer. Why? A proper rehearsal is missing — and what a rehearsal lacks most is a sparring partner.

Senior sellers have no time to play the customer, and they play it badly (they know their own rebuttals too well to sound like a genuinely difficult buyer); colleagues role-playing each other drift into going through the motions; and you cannot book real customers for practice. So newcomers practice on real customers — the approach where everyone pays the price.

This article lays out one concrete landing: let AI play the customer inside a 3D training scene you deployed yourself. First, the division of labor — the platform provides the body and presence; the brain playing the customer is the LLM you connect. Implemented versus planned are kept separate.

I. Why rehearsals are hard to organize

A competent reception rehearsal needs four things:

1. A scene: a believable reception environment — a store, a showroom, an equipment site — not small talk in a meeting room;
2. An opponent: a "customer" who probes, pushes back, and suddenly changes topics — not a colleague reading a script;
3. Presence: practicing in a space with room, exhibits, and walking — not typing at a screen;
4. Review material: what was said, where it stalled — recorded and replayable.

The first three are why rehearsals never get organized: real scenes cost money, and human stand-ins cost time. The fourth used to rely on audio recordings and a mentor's memory.

II. What the platform provides, and where the brain lives

Pin down the division of labor first — this is where evaluations of such solutions get most confused:

In other words: the platform's agent only provides "a character that can walk, talk, and be seen in the scene." How it responds to the trainee's pitch is entirely up to your client and model. Which also means rehearsal scripts, difficulty levels, and scoring criteria all stay in your hands — no platform in the loop.

III. What landing looks like

Before integration: build a simplified reception scene in the editor (one store or one booth is enough; don't overbuild); write prompts for the customer roles — budget-sensitive, comparison-shopping, layman-questioning — one per typical customer archetype.

During integration: three steps — the discovery file (.well-known/virtual-world-agent.json), exchanging the Key for a 15-minute token, connecting to the dedicated channel. The trainee walks into the 3D scene with their own account; the AI-played customer stands behind the counter. The trainee opens, the "customer" responds. When the trainee stalls, the scene just stays there — the stall itself is part of the rehearsal.

After integration: pull the conversation logs for review — which line lost the "customer's" interest, which pricing question went unanswered. Swap in another customer prompt and run it again. Every trainee's logs sit in your own database, so mentors can replay session by person.

IV. Honest limits: what it cannot train

One-sentence positioning: it is a practice opponent who is always available, never tires, and always shows up — not a replacement for your mentors. Once the fundamentals are drilled, the real floor awaits.

V. What to prepare

• A simplified 3D scene (built in the editor; business staff can join);
• Customer-role prompts (the bulk of the work — they decide rehearsal quality);
• A code-capable AI client (the repo ships a zero-dependency Node example that runs the full pipeline);
• Your own server (Node 18+ and a database).

VI. What it does not suit

• Anyone expecting AI to fully replace mentoring and real-customer contact — it drills the fundamentals, not everything;
• Industries where training hinges on body language and in-person presence — text-based practice cannot cover those;
• Teams with nobody to write scripts and prompts — without good scripts, rehearsal is just chatting.

VII. FAQ

Q: How is this different from a regular web-based AI role-play?

A: Two ways. Scene — the trainee practices inside a walkable 3D space, walking up to exhibits and facing the "customer," closer to real reception than a chat box. Data — conversation logs persist to your own server in real time, so trainee practice data never leaves your environment.

Q: Can one AI drill several trainees at once?

A: Technically you can run multiple agents in one world (measured: about 1 KB/s per agent; one hundred in a world total about 0.8 Mbps, server load around 1% of a single core), but rehearsal dialogue is usually one-on-one — several trainees shouting around one "customer" degrades quality. One practice slot per person is the sensible setup.

Q: Does the trainee know the counterpart is AI?

A: Yes. AI identity is always explicit — an entry notice and an overhead prefix. This is a product red line. In a rehearsal scene it is actually appropriate: be clear this is practice, so that what is practiced is real.

VIII. Source code and repositories

All three mirrors hold identical content; the first two are faster for visitors in mainland China. The repos include deployment guides and a demo entry.

• Gitee (faster in mainland China): https://gitee.com/miduoxinxijeji/miduo.git
• GitCode (mirror): https://gitcode.com/qq_35054471/virtual-world
• GitHub: https://github.com/miduo100/3d-virtual-world

About Genesis

Genesis is a self-hosted 3D virtual world system built on Three.js + WebGL, helping individuals and businesses build their own 3D spaces. Accessible directly from a browser, compatible with both PC and mobile, it supports multiplayer online, federated teleportation, a shop system, and Agent integration—where an AI can enter your world as an embodied character. Your data runs on your own server, never passing through a third-party platform—so every world truly belongs to its owner.

Want a sparring partner who is always available for your new hires? Genesis (the Genesis Virtual World CRM System) is a Three.js 3D virtual-world base deployed on your own server — you build the scene, you connect the brain, you keep the logs, and AI steps into your training scene as a humanoid sparring partner. The official site (search "Genesis Virtual World CRM") has a demo world you can walk through.

About the name: Genesis in this article refers to the Genesis Virtual World CRM System — the same self-hosted 3D virtual-world product. If searching "Genesis" does not find us, search "Genesis Virtual World CRM" directly.