Agent Cost Strategy

A tiered model selection framework for multi-agent workflows. Use the cheapest model that can reliably do the job.

The Tiers

Tier	Use For	Examples
Fast/Cheap	Sub-agents, background workers, iterative fixes, well-defined single-step tasks	Claude Haiku, GPT-4o-mini, Gemini Flash
Mid-tier	Main dialogue, day-to-day assistance, moderate complexity tasks	Claude Sonnet, GPT-4o, Gemini Pro
Powerful	Architecture decisions, deep code reviews, hard problems, when cheaper models fail twice	Claude Opus, GPT-4.5, Gemini Ultra

Decision Rules

Use Fast/Cheap when:

Task is well-scoped and single-step
Input/output is straightforward (fix this test, summarize this, run this command)
It's a background/automated task with no user interaction
You're running many parallel sub-agents

Use Mid-tier when:

Conversational context matters
Task requires moderate reasoning or multi-step thinking
This is the default for your main assistant session

Use Powerful when:

Cheaper models have failed 2+ times on the same problem
Making high-stakes architectural decisions
Deep code review or security audit
The cost of a wrong answer exceeds the cost of the model

Sub-Agent Pattern

When delegating to a sub-agent, default to the cheapest model that fits the task:

Task type               → Model tier
─────────────────────────────────────
Fix failing tests       → Fast/Cheap
Write boilerplate       → Fast/Cheap
Research/search         → Fast/Cheap
Cron/scheduled tasks    → Fast/Cheap (always)
Short replies (hi/ok)   → Fast/Cheap (always)
Build new feature       → Mid-tier
Review PR               → Mid-tier
Architecture            → Powerful
Stuck after 2 tries     → Escalate up one tier

Heartbeat Interval

Set heartbeat to 55 minutes (not 30) when using Anthropic API keys. This keeps the prompt cache warm just under the 1-hour TTL — every heartbeat pays cheap cache-read rates instead of re-writing the full cache.

"heartbeat": { "every": "55m" }

Communication Pattern Rule

Short conversational messages (hi, thanks, ok, sure, got it, yes, no) should always use Fast/Cheap models. Never burn Sonnet or Powerful on one-word acknowledgments.

Cache Optimization

Prompt caching can cut costs by 80-90% on repeated context. See references/cache-optimization.md for patterns.

Tracking

Monitor spend by checking your provider's usage dashboard regularly. Signs you're over-spending:

Running Powerful models on tasks Fast/Cheap can handle
No caching on repeated system prompts
Spawning sub-agents without a model tier strategy
Heartbeat set to 30min (re-writes cache every time)