How to save up to 70% of tokens with LLMs?

2026-06-21 · Source: LLM on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, medium

Summary

An analysis of LLM token usage in AI Agentic workflows reveals that optimizing database schema representation within prompts can yield substantial savings. The author achieved an 84% token reduction in a specific project. For a test "Dummy project" with 20 interconnected database tables, converting the schema from JSON (48144 characters, 9106 tokens using the cl100k_base tokenizer) to TypeScript (10412 characters, 2758 tokens) resulted in a 69.71% reduction, saving 6,348 tokens per request. This optimization not only decreases context costs and processing latency but also improved LLM accuracy. The "syntax tax" of JSON Schema, characterized by verbose `required` arrays, nested enums, explicit type annotations, and token fragmentation, makes it less efficient than TypeScript's concise syntax.

Key takeaway

For AI Engineers optimizing LLM agentic workflows, consider replacing verbose JSON Schema representations of database structures with TypeScript interfaces in system prompts. This approach can reduce token consumption by up to 69.71% and improve model accuracy, significantly lowering operational costs and latency. You should use TypeScript for general text formatting and extraction tasks, reserving JSON Schema for scenarios demanding strict API grammar validation.

Key insights

Representing database schemas as TypeScript interfaces in LLM prompts significantly reduces token usage and improves accuracy.

Principles

• Concise syntax reduces token consumption.
• Fewer tokens can improve LLM attention/accuracy.
• Context window optimization is crucial for agentic systems.

Method

Convert Zod schemas to TypeScript interfaces using `zod-to-ts`, inject the resulting string into the system prompt, and validate LLM output with Zod.

In practice

• Use TypeScript for cost-sensitive LLM text formatting.
• Employ JSON Schema for strict API grammar validation.
• Implement Zod for runtime validation of LLM outputs.

Topics

• LLM Token Optimization
• AI Agentic Workflows
• Database Schema Representation
• TypeScript Interfaces
• JSON Schema
• Zod Validation

Best for: Prompt Engineer, AI Engineer, MLOps Engineer

Related on AIssential

• Notation Matters: A Benchmark Study of Token-Optimized Formats in Agentic AI Systems — Token-optimized formats like TRON significantly reduce LLM token consumption in agentic systems with acceptable accuracy trade-offs.
• Why MCP Is Eating Your Context Window (and How Apideck CLI Fixes It) — Verbose JSON tool definitions in MCP rapidly consume LLM context windows, limiting conversational depth.
• Cross-Lingual Token Arbitrage: Optimizing Code Agent Context Windows via Local LLM Preprocessing — Proactive, edge-side prompt rewriting with a local LLM significantly reduces token costs for coding agents.
• TOON: Beyond JSON for LLMs — TOON offers a token-efficient structured data format specifically designed to optimize LLM context window usage over traditional JSON.
• Remember, Don't Re-read: Stateful ReAct Agents for Token-Efficient Autonomous Experimentation — Stateful ReAct agents significantly reduce token costs in autonomous LLM experimentation by maintaining context across iterations.
• rtk-ai / rtk — RTK significantly reduces LLM token consumption by filtering and compressing command outputs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by LLM on Medium.