RepReady

Enterprise sales teams rehearse deals on live prospects or not at all. Live roleplay needs two senior people in a room; most teams skip it. RepReady gives reps unlimited reps against an AI buyer that keeps a consistent persona, memory, and objection history across sessions.

A roleplay agent that agrees with everything teaches nothing. The buyer had to push back believably, remember what it conceded last week, and escalate its own objections as the rep improved. That is a memory problem dressed up as a persona problem, and it is the part that separates demos from drills.

The published platform separates persona, context and memory, reasoning, evaluation, and performance intelligence into five layers. The rep conversation runs through an orchestration loop with two agents in the loop; buyer memory is written per session and retrieved at the next session start. Evaluation runs on a rubric per deal stage, and debriefs cite the exact turn where the rep lost the objection.

• Buyer personas are structured data plus a presentation layer, not free text.
• Memory writes are gated by the orchestration loop, never by the model.
• The debrief generator is evaluated against human-written debriefs before release.

Early buyer personas drifted toward friendliness within two sessions: the persona prompt degraded as conversation history grew. Fix: move persona constraints into the memory layer with higher priority than conversation history, and assert persona consistency in the evaluation layer. A second miss: reps gamed the quality score by switching topics; the rubric now weights objection handling over topic coverage.

• Persona is a constraint system, not a prompt.
• Next: publish performance metrics once disclosure clears, and ship the multi-buyer negotiation mode.