💡 Workflow Experience Replay: Interactive Avatar-Guided Demo Mode for Team Onboarding

[github-actions[bot]]

Summary

Record complete avatar-guided workflow sessions as interactive replays that colleagues can watch, experience, and then launch themselves. Unlike session summaries (#231 — static output sharing) or decision replay (#423 — audit narration for explainability), this captures the full interactive experience — avatar speech, overlay transitions, user decision points — so viewers feel what using TalkTerm is like before they try it. The avatar introduces the demo: "This is how Sarah ran her brainstorming session last week. Watch along, and when you're ready, I'll help you start your own."

Market Signal

88% of AI agent pilots never reach production deployment (IBM 2026, Arcade.dev) — the primary barrier is organizational, not technical. Employees resist AI systems that change established processes (Sana Labs). Demo/replay directly addresses the "show, don't tell" adoption gap. Loom proved that async video sharing drives tool adoption within teams; interactive replay takes this further by letting the viewer experience the tool's avatar interaction, not just watch a screen recording. No competitor offers interactive AI workflow replay — this is a white-space opportunity.

User Signal

TalkTerm's PRD success criteria requires "3+ non-technical users completing workflows independently within 60 days of launch." Discussion #231 (Shareable Session Summaries) covers sharing output artifacts but not the experience of producing them. Discussion #423 (Avatar Decision Replay) covers post-hoc explainability — walking through decisions for auditing — but not onboarding or adoption replay. Discussion #245 (Guided Interactive Onboarding, approved and in development) addresses first-launch onboarding but not peer-to-peer adoption. The adoption gap is: Sarah completes her brainstorming, loves it, but can't show her colleague what the experience felt like beyond sharing the output document.

Technical Opportunity

TalkTerm's event-based IPC architecture already generates structured events (SDK messages, avatar state transitions, overlay panel opens/closes, user interactions) that flow through well-defined channels. Recording these events requires no new infrastructure — just persisting the event stream with timestamps to a session replay file.

MVP Implementation Path

Phase	Capability	Complexity
Phase 1: Record	Capture IPC event stream + TTS audio as timestamped timeline	Low — event logging extension
Phase 2: Playback	Replay avatar speech and overlay transitions sequentially	Medium — playback engine driving existing UI components
Phase 3: Interactive	Pause at decision points, let viewer make choices	Medium — branch points in replay
Phase 4: Export	Self-contained replay file for sharing across TalkTerm instances	Low — serialization

The engineering cost is amortized across the onboarding epic (#245) which already requires understanding event sequencing for guided first-session experiences. The same event stream schema serves both guided onboarding and experience replay.

Assessment

Dimension	Score	Rationale
Feasibility	med	Event-based architecture supports recording naturally; playback engine is new but bounded
Impact	high	Directly addresses the #1 barrier to AI agent adoption (organizational resistance)
Urgency	med	Important for team adoption post-MVP; not blocking individual user validation

Adversarial Review

Strongest objection: Building a full replay engine is significant engineering effort for a feature that's only valuable after multiple users adopt — a chicken-and-egg problem. Why invest before proving individual user value?

Rebuttal: MVP replay does not require full state reconstruction — Phase 1-2 (record event stream + playback avatar speech/overlays) is a guided tour, not a pixel-perfect simulation. This is far more compelling than screen recordings for showing colleagues what AI-assisted workflows feel like. Sensitive data stripping uses the same classification as the audit trail (FR27-29). Most importantly, TalkTerm's success metric requires multi-user adoption within 60 days — replay mode is a force multiplier for that goal, not a nice-to-have. The investment in event stream recording also benefits debugging, QA, and the approved onboarding epic (#245).

Suggested Next Step

Define the event stream schema for session recording: which IPC events to capture, what metadata to include, and the replay file format. Prototype recording an event stream during a BMAD brainstorming workflow and replaying it as sequential avatar narration with overlay transitions.