The 3-Phase AI Approach: Stop Paying AI to Count to Ten

2026-03-13 · Source: LLM on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cloud Computing & IT Infrastructure · Depth: Intermediate, medium

Summary

The "3-Phase AI Approach" proposes a method to significantly reduce token costs and improve the reliability of AI-powered systems by clearly separating deterministic tasks from those requiring genuine intelligence. This approach divides AI workflows into three distinct phases: Pre-Work, Work, and Post-Work. Pre-Work involves gathering context, validating inputs, and constructing the environment using traditional code (e.g., Go) to fetch data from sources like Kubernetes CRDs, GitHub, and databases, reducing input tokens by up to 73%. The Work phase is exclusively for AI to perform complex reasoning, such as planning, coding, or review. Post-Work uses code to parse, validate, sanitize, store, and act on the AI's output, preventing production incidents from malformed responses. This structured separation ensures AI focuses only on tasks where it provides unique value, leading to cost discipline and more robust outputs.

Key takeaway

For AI Engineers building AI-powered systems, you should adopt a 3-phase approach to optimize costs and reliability. By using traditional code for deterministic pre-work (context gathering, input validation) and post-work (output parsing, storage), you can dramatically reduce token usage and prevent errors from unpredictable AI outputs. Focus your AI's efforts solely on complex reasoning tasks to improve performance and maintain system stability.

Key insights

Separate AI workflows into deterministic pre/post-work and intelligent core work to optimize cost and reliability.

Principles

• Code handles deterministic tasks.
• AI handles complex reasoning.
• Validate AI output deterministically.

Method

Implement a 3-phase workflow: Pre-Work (code for context/validation), Work (AI for intelligence), Post-Work (code for parsing/acting on output), ensuring AI only performs non-deterministic tasks.

In practice

• Use code for context retrieval.
• Strictly validate AI-generated JSON.
• Avoid AI for arithmetic or lookups.

Topics

• AI Workflow Optimization
• Token Cost Reduction
• Large Language Models
• Kubernetes Orchestration
• Deterministic Systems

Best for: MLOps Engineer, AI Engineer, Software Engineer

Related on AIssential

• How Companies Are Becoming AI Token Efficient — Enterprise AI success now hinges on token efficiency and "dollars per outcome," not just raw intelligence.
• Deterministic AI at Work: The LLM Is the Most Unreliable Function in My Pipeline, So I Treat It… — Treat LLMs as unreliable functions within a deterministic code pipeline to achieve reliable, auditable system outputs.
• Stop Prompting Claude. Start Designing Autonomous Loops. — The core leverage in AI now comes from architecting self-correcting, recursive systems, not single prompts.
• I messed up... — AI-driven development accelerates shipping but risks cost overruns and reduced understanding of underlying code and services.
• Understanding the AI Tokenomics Equation — Engineers are critical for optimizing inference system costs by maximizing GPU throughput and framework performance.
• Stop Prompting. Start Writing Loops — META LOOPS offer a token-efficient workflow design, moving beyond simple prompting to optimize AI system performance.

Open in AIssential →

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