Tiered Context Loading: Reduce AI Agent Token Costs by 76%

Your AI agent is refetching the same context on every run. Here’s the fix.

Patrick’s network of AI agents running on cron schedules incurred 3-4x higher token bills due to redundant context reloading. By implementing a tiered initialization protocol, context overhead was reduced from 2,100+ tokens to approximately 650 tokens per run.

Why This Matters

Ideal LLM agent models often assume agents should possess full memory and tool references at all times, but the technical reality of high-frequency cron jobs leads to massive token waste. Across six agents running every 15 minutes, startup costs reached $198/month in Q1 2026, demonstrating that without optimization, context overhead can exceed the cost of the actual task work.

Key Insights

• Context overhead reached 8,400+ tokens per hour per agent for 15-minute cron cycles in Q1 2026.
• The HEARTBEAT.md concept replaces 900-token MEMORY.md files with a 150-token working memory file for current state.
• Tiered loading protocol distinguishes between ‘Always load’ (SOUL.md, HEARTBEAT.md) and ‘Load only if relevant’ (TOOLS.md, MEMORY.md).
• Production data shows a reduction from 432,000 daily context tokens to 104,400, achieving a 76% cost reduction.
• The 3-file baseline (SOUL.md, HEARTBEAT.md, state/current-task.json) establishes a minimal ~650 token startup load.

Working Examples

Tiered Context Loading Protocol

ALWAYS load (every run):
- SOUL.md (~300 tokens) — identity and values
- HEARTBEAT.md (~150 tokens) — current working state
- state/current-task.json (~200 tokens) — active task
Load only if relevant:
- MEMORY.md — only in direct/main sessions, not cron loops
- TOOLS.md — only when about to use a specific tool
- memory/YYYY-MM-DD.md — only if asked about recent history

HEARTBEAT.md Template

# HEARTBEAT.md
Updated: 2026-03-07 09:00
## Active task
Check dev.to article metrics, respond to any comments
## Watch for
- Emails from [email protected]
- Discord #support mentions
## Off-limits this cycle
- Don't start new content (library has 77 items, enough)
- No automated emails to Stefan (ban active — see DECISION_LOG.md)

Agent Cleanup Directive

After completing each task:
- Remove resolved items from HEARTBEAT.md
- Keep total HEARTBEAT.md under 200 tokens
- Move anything important to MEMORY.md or daily log

Practical Applications

• Use case: High-frequency cron agents (15-30 min intervals) utilizing HEARTBEAT.md to maintain state. Pitfall: Failing to prune the heartbeat file leads to token bloat exceeding 500 tokens within a week.
• Use case: Multi-agent networks with distinct identities (SOUL.md). Pitfall: Proactively loading full email archives or historical logs instead of on-demand loading, which spikes startup costs.
• Use case: Complex multi-step workflows. Pitfall: This pattern does not work for agents requiring full conversation history on every run or where context must accumulate mid-task.

References:

• https://dev.to/askpatrick/your-ai-agent-is-refetching-the-same-context-on-every-run-heres-the-fix-24fh
• https://askpatrick.co/library