tachigo AI Native Product Roadmap

Purpose

This document captures Track B: AI-native product direction for tachigo.

It complements Discussion #877, which remains the short-term source of truth for Track A: AI-native repo / workflow foundation.

This is a planning reference, not an implementation source of truth. Before implementation, each section must be converted into issues, specs, acceptance criteria, and small PRs.

Relationship to Discussion #877

Recommended split:

Track A: AI-native repo / workflow foundation
- Source of truth: Discussion #877
- Focus: Dev Portal, Wiki, AI Agent Start Here, domain cards, issue templates, PR gates, review / merge guardrails
- Timing: short term

Track B: AI-native product roadmap
- Source of truth: this document, after scope is accepted
- Focus: Campaign Copilot, viewer insights, economy advisor, recommendation feedback loop
- Timing: medium term

Short-term priority:

Make tachigo AI-readable and AI-safe before making tachigo product features AI-powered.

Product Positioning

Recommended positioning:

tachigo is an AI-native creator rewards OS for Twitch streamers, agencies, and viewers.

AI should help streamers and agencies:

• understand viewer behavior
• design reward campaigns
• manage token economy health
• detect suspicious reward farming
• turn dashboard data into operational actions

AI Native Definition

For tachigo, AI Native means:

AI is embedded into creator operations, reward design, viewer insight, and economy optimization.

The product should evolve from:

viewer watches stream → heartbeat → points increase → token claim / spend

Into:

viewer activity → AI insight / recommendation → streamer review → deterministic backend execution → feedback loop

Core principle:

AI is not the authority.
AI is the operating layer for recommendations, insights, and optimization.

Recommended First Product Feature

Do not start with a generic chatbot.

The first AI-native product feature should be embedded into a concrete dashboard workflow:

Dashboard Campaign Copilot

Recommended workflow:

Create Campaign
  → streamer inputs goal
  → AI generates campaign draft
  → backend calculates cost / risk
  → streamer edits
  → streamer confirms
  → campaign is saved / published

The first version should generate a campaign draft, not publish it automatically.

Detailed safety, privacy, and execution rules live in ai-copilot-principles.md.

Product Feature Phases

Phase 1: Dashboard Campaign Copilot

Help streamers and agencies create reward campaigns from goals.

Example draft components:

- Watch Streak Mission
- Chat Burst Mission
- TACHI Sink
- New Viewer Bonus
- Risk controls

Phase 2: Viewer Segment Insight

Help streamers and agencies understand viewer behavior and convert analytics into actions.

Example segments:

- Core fans
- Silent loyal viewers
- Reward hunters
- New viewers
- Dormant viewers

Phase 3: Token Economy Advisor

Help streamers and agencies manage reward economy health.

Metrics to analyze:

- weekly point issuance
- weekly point spending
- sink ratio
- spendable balance growth
- whale concentration
- reward farming pattern
- inactive token accumulation

Differentiation:

Generic tool: helps issue points.
tachigo: helps design, monitor, and adjust the creator reward economy.

Phase 4: Recommendation Feedback Loop

The product should track:

What did AI recommend?
Did the streamer accept it?
Why was it rejected?
Did accepted recommendations improve metrics?

Candidate data areas:

ai_recommendations
ai_recommendation_feedback
ai_insight_snapshots

Backend Direction

Add an independent AI bounded context only after Track A has made the repo AI-readable and AI-safe.

Boundary rule:

AI reads multiple domains and produces recommendations.
It does not own ledger, auth, wallet, or campaign execution.

Recommended ownership:

points service owns points.
agency service owns agency/channel permissions.
campaign service owns campaign lifecycle.
AI service owns recommendation lifecycle.

Suggested Roadmap

Phase 0: AI-native repo / workflow foundation from #877
Phase 1: Product roadmap and AI copilot principles docs
Phase 2: Campaign Copilot MVP with deterministic TemplateProvider
Phase 3: LLM-backed Campaign Copilot with redaction and schema validation
Phase 4: Viewer Insight
Phase 5: Economy Advisor
Phase 6: Semi-autonomous creator operations with human approval

Suggested PR Breakdown

PR 1: Track B product roadmap and copilot principles docs
PR 2: AI recommendation domain skeleton
PR 3: AI recommendation persistence
PR 4: Campaign Copilot API with deterministic TemplateProvider
PR 5: Dashboard Campaign Copilot UI
PR 6: LLM provider adapter
PR 7: Viewer Segment Insight
PR 8: Economy Advisor

Each implementation PR should have its own source issue, explicit non-goals, acceptance criteria, and test plan.

North Star Metrics

Measure product impact, not whether AI output looks smart.

Streamer: campaign creation time decreases, accepted AI recommendation ratio increases.
Viewer: repeat viewer ratio, watch retention, reward redemption, and chat participation improve.
Economy: sink ratio improves, token inflation risk and suspicious farming decrease.
AI loop: accepted recommendation → measurable product metric improvement.

Non-goals

This roadmap does not require the current PR to:

• implement backend AI services
• add migrations
• add dashboard UI
• connect to an LLM provider
• modify points or token ledger behavior
• change wallet or claim logic
• add dependencies
• add autonomous campaign publishing
• replace Discussion #877 as the Track A source of truth

Conclusion

The short-term priority remains Track A from Discussion #877:

Make the repo AI-readable and AI-safe first.

Then Track B can proceed as an AI-native creator rewards OS, starting with Dashboard Campaign Copilot.