AI Tool Ecosystem Optimization Strategy
Comprehensive strategy for routing development tasks across a multi-tier AI tool stack, optimizing for cost, capability, and security boundaries. Defines the operational playbook for Tiered AI Task Routing and integrates with Integrated Operations Strategy.
The Stack
| Subscription | Monthly Cost | Primary Role | Rate Limits |
|---|---|---|---|
| Claude Code Max 20x | ~$200 | Orchestrator, architecture, security | 20x standard usage |
| Google AI Ultra | ~$250 | Gemini CLI, Deep Research, Jules, 30TB storage | Highest Gemini limits |
| OpenCode + Oh My OpenCode | Free (OSS) | Multi-model routing, agent orchestration | Provider-limited |
| Aider | Free (OSS) | Git-integrated pair programming | Provider-limited |
| Gemini CLI | Free tier | Lightweight coding agent | 60 req/min, 1K/day |
| Qwen Code | Free (OSS) | Boilerplate, local-capable | Unlimited local |
| GitHub Copilot | Included | Inline autocomplete | Standard |
Core Philosophy: Task-Model Fit
The single most expensive mistake is using a $200/month orchestrator model to rename variables. The single most dangerous mistake is delegating security-critical work to an unvetted free tool.
Routing Hierarchy
- Tier 1 (Claude Code Max): Architecture, security, ops-core, orchestration — never delegated
- Tier 2 (Gemini CLI / Google AI Ultra): Deep Research, docs, large-context analysis, Jules async tasks
- Tier 3 (OpenCode + OMOC): Multi-model routing, parallel background agents, UI work, bulk refactors
- Tier 4 (Aider / Qwen / Free): Boilerplate, single-file edits, exploration, spikes
Detailed Tool Roles
Claude Code Max — The Orchestrator
Reserve for work touching ops-core, cross-zone dependencies, auth/authorization, and production-shipping code. Pair with Plane tickets labeled cc-arch or cc-security. Use /compact and context management to keep sessions lean.
Anti-patterns: Fixing README typos, documentation generation (Gemini handles equally well), idle open sessions.
Google AI Ultra — Research Engine
- Deep Research: Multi-page reports before starting any new project/technology evaluation
- Gemini CLI: Documentation generation, code explanation, large-context tasks
- Jules: Async background tasks (dependency updates, test scaffolding)
- 30TB Storage: Canonical backup and artifact store
OpenCode + Oh My OpenCode — Multi-Model Router
Model-agnostic CLI with specialized agents routing to best-fit models. Free built-in models handle exploration at zero cost. Google AI Ultra feeds Gemini models for heavier work.
Key agents: Sisyphus (structured task completion), Prometheus (planner mode), Atlas (strategic reasoning), Librarian (documentation retrieval).
Aider — Git-Native Pair Programming
Multi-file coordinated edits with automatic git commits. Sweet spot: Aider + Gemini API = high-quality refactoring at low cost with automatic git history.
Qwen Code — Free Heavy Lifting
Boilerplate, test scaffolding, grunt work. Running locally via Ollama provides zero-cost, fully private coding assistance for sensitive exploration.
Workflow Examples
New Feature Development
- Research (Cost: $0): Gemini CLI + Deep Research
- Architecture (Cost: Claude usage): Design, interfaces, security boundaries
- Implementation (Cost: Gemini API via OpenCode): Sisyphus agent with parallel background tasks
- Review & Harden (Cost: Claude usage): Security review, cross-zone audit
Cost profile: ~60% free tools, ~25% Gemini, ~15% Claude Code.
Knowledge Harvesting Pipeline
Deep Research → Gemini CLI formats → Claude Code reviews → Aider commits → Google Drive archives. ~20 minutes vs 2-4 hours manual.
Daily Usage Allocation
| Time Block | Primary Tool | Task Type | Cost Tier |
|---|---|---|---|
| Morning standup | Gemini CLI (free) | Review, plan | Free |
| Deep feature work | Claude Code Max | Architecture, security | Premium |
| Broad implementation | OpenCode + OMOC | UI, parallel tasks | Mixed |
| Documentation | Gemini CLI / Deep Research | Docs, reports | Included |
| Refactoring | Aider + Gemini API | Multi-file changes | Low cost |
| Quick fixes | OpenCode (free model) | Typos, config | Free |
| End of day | Gemini CLI | Knowledge capture | Free |
Cost Optimization Summary
| Strategy | Estimated Savings |
|---|---|
| Route docs to Gemini (not Claude) | 15-20% Claude usage |
| OpenCode free models for exploration | 10-15% paid API calls |
| Aider + Gemini for refactoring | 20-30% vs Claude |
| Deep Research replaces manual research | 2-4 hrs/week saved |
| Parallel background agents | 40-60% time reduction on bulk |
| Local Qwen for sensitive exploration | 100% cost reduction for spikes |
Bottom line: ~$450/month subscription, 3-4x output vs running everything through Claude alone.
Security Boundaries (Non-Negotiable)
- ops-core governance and security repos — Claude Code Max only
- Auth, authorization, cryptographic code — Claude Code Max only
- No sensitive code to free/unvetted models
- API keys and secrets — never through cloud-hosted free models
- Security review is always the final gate