Mono-Repo Consolidation: Executive Summary

TiDB Agentic Engineering AI-First Initiative

The Vision

Build a mono-repo where AI can autonomously:

• Design system architecture
• Develop features end-to-end
• Test and validate changes
• Deploy and monitor services
• Iterate based on outcomes

This is not just code consolidation. This is building the foundation for General Relativity: AI owns the full engineering lifecycle.

The Problem

Current State: 400+ Repositories, ~39GB
├── Products: TiDB, TiDB Next-Gen
├── Platform: TiDB Cloud SaaS
├── DevOps: Operations tools
├── Forks: Third-party dependencies
└── Abandoned: Unused projects

Issues:
❌ AI cannot see full system context
❌ Cross-repo optimization is impossible
❌ Human coordination overhead scales with repo count
❌ Dependency hell across repos
❌ Inconsistent tooling and practices

The Solution

Target State: 1 Unified Mono-Repo
├── AI-readable structure
├── AI-optimizable boundaries
├── Automated build/test/deploy
├── Clear ownership (CODEOWNERS)
└── Trunk-based development

Google’s Playbook (2 Billion LOC Proven)

Key Insight: If monorepo works for Google at 2B LOC with 25K engineers, it can work for us.

Our AI Advantage

Google built their system before AI was mainstream. We have a unique advantage:

Google (Human-Centric Automation)

Humans: Write code, review, fix dependencies, deploy
Automation: Formatting, dependency updates, builds, tests

Us (AI-First)

AI Agents: Write code, review, fix dependencies, optimize builds, deploy decisions
Humans: Define problems, set priorities, review architecture, handle edge cases

We’re not just matching Google. We’re going beyond.

Three-Layer AI Development Model

┌─────────────────────────────────────────────────────────────┐
│                    AI Capability Layers                     │
├─────────────────────────────────────────────────────────────┤
│  Micro   │  Skills, MCP, Tools           │ Current state    │
│          │  (Efficiency in existing)     │                  │
├─────────────────────────────────────────────────────────────┤
│  Meso    │  Feature lifecycle            │ Phase 4.2        │
│          │  (AI drives design→deploy)    │                  │
├─────────────────────────────────────────────────────────────┤
│  Macro   │  System architecture          │ Phase 4.3        │
│          │  (AI reorganizes everything)  │                  │
├─────────────────────────────────────────────────────────────┤
│  General │  AI owns everything           │ End state        │
│  Relativity                            │                  │
└─────────────────────────────────────────────────────────────┘

Project Phases

Phase 1: Repository Analysis (Week 1-2)

400+ AI Agents analyze all repos

Deliverable: repo-analysis-report.md

Phase 2: Mono-Repo Design (Week 2-3)

Infrastructure setup

mono-repo/
├── products/          # TiDB, TiDB Next-Gen
├── platform/          # Cloud SaaS
├── devops/            # Operations
├── libs/              # Shared libraries
├── tools/             # Build/dev tools
└── infra/             # Infrastructure

Key Decisions:

• Build system (Bazel vs Turborepo vs Nx)
• CODEOWNERS structure
• CI/CD path-based triggering
• Branching model (trunk-based)

Deliverables: mono-repo-structure.md, codeowners-template.md, build-system-evaluation.md

Phase 3: Pilot Migration (Week 3-4)

10-20 repos (P0 priority)

Deliverable: migration-runbook.md (refined from pilot)

Phase 4: Bulk Migration (Week 4-8)

Remaining ~380 repos in batches

Phase 5: AI Enablement (Week 8+)

Closed-loop development

Deliverable: ai-dev-loop-spec.md, ai-first-methodology.md

Success Metrics

Resource Requirements

Infrastructure

Tooling

• Build System: Bazel / Turborepo / Nx
• Code Search: Sourcegraph / Zoekt
• CI/CD: GitHub Actions / GitLab CI
• Agent Framework: Custom (Python/Go)

Team

• Project Lead: 1 FTE
• Build/Infra Engineer: 1-2 FTE
• AI/ML Engineer: 1-2 FTE
• Team Representatives: 0.2 FTE each (for migration decisions)

Risks & Mitigation

Open Questions (Need Answers)

1. Tech Stack: What languages/frameworks are in the 400 repos?

   • Determines build system choice (Bazel vs Turborepo vs Nx)

2. Current CI/CD: What’s the existing pipeline?

   • Affects migration complexity

3. Team Structure: How many engineers? How organized?

   • Affects CODEOWNERS design

4. Deployment: How are services currently deployed?

   • Affects infra design

5. Agent Hosting: Where will 400 agents run?

   • Local cluster? Cloud? Hybrid?

Next Steps (Planning Phase: 1-2 Days)

Day 1: Analysis Framework

• Set up distributed agent infrastructure
• Define analysis metrics and scoring
• Create repo inventory (list all 400 repos)
• Run pilot analysis on 10 repos

Day 2: Mono-Repo Design

• Finalize directory structure
• Design build system architecture
• Plan migration tooling
• Create detailed migration runbook

Deliverables

• repo-analysis-report.md
• mono-repo-structure.md
• migration-runbook.md
• ai-dev-loop-spec.md
• ai-first-methodology.md
• ai-capability-maturity.md
• google-monorepo-lessons.md ✅ DONE
• codeowners-template.md ✅ DONE
• build-system-evaluation.md

Conclusion

This project is not just about consolidating code. It’s about:

1. Building the foundation for AI to own the full engineering lifecycle
2. Learning from Google’s playbook (2B LOC proven)
3. Going beyond Google with AI-first decision automation
4. Enabling Agentic Engineering at scale

The goal is not to help humans do AI work. The goal is to have AI do the work, and humans define what matters.

Prepared for: TiDB Agentic Engineering AI-First Initiative Last updated: Planning Phase