My AI Coding Workflow

2026-06-03 · Source: Refactoring · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

The author details an updated AI coding workflow for the Tolaria project, highlighting significant productivity gains and operational efficiency. Since its April 22nd release, Tolaria has achieved impressive metrics, including 6 open issues versus 417 closed, 14 open PRs against 306 closed, a 1-day average bug fix time, and a 99.1% crash-free rate. The codebase now stands at 150K LOC plus ~100K LOC of tests, with the author contributing ~1000 commits while working approximately 2 hours daily. Key workflow tweaks include a "Guides, Gates, and Guards" mental model for AI instructions, a transition from Claude to Codex, which reduced monthly costs by over 90%, and a focus on continuous improvement.

Key takeaway

For AI Engineers optimizing their coding workflows, adopting a structured "Guides, Gates, and Guards" approach to AI instructions can significantly enhance code quality and development velocity. You should consider defining explicit rules for TDD, UI components, and localization within your AI agent's core instructions, potentially reducing operational costs by switching models like from Claude to Codex.

Key insights

The author's AI coding workflow leverages a "Guides, Gates, and Guards" model, shifting to Codex for efficiency and cost reduction.

Principles

• AI instructions can be categorized as Guides, Gates, and Guards for structured application.
• Consolidating core instructions into a single AGENTS.md file can simplify AI agent management.
• Quantitative measurement of code quality (e.g., CodeScene, Codacy) should be integrated into AI-driven development.

Method

The proposed AI instruction method involves sequential application of Guides (initial context), Gates (in-progress steering), and Guards (fallback procedures) to improve AI output quality and enforce development standards.

In practice

• Define core AI instructions for TDD, Boy Scout Rule, UI standards, and localization in AGENTS.md.
• Utilize shadcn/ui components and Lara CLI for consistent UI and localization with AI agents.
• Consider switching LLMs (e.g., Claude to Codex) for potential cost savings and improved performance.

Topics

• AI Coding Workflow
• AI Agent Instructions
• Large Language Models
• Codex
• Software Development Metrics
• Test-Driven Development
• Code Quality

Code references

• refactoringhq/tolaria

Best for: AI Engineer, Software Engineer

Related on AIssential

• Code smells for AI agents: Q&A with Eno Reyes of Factory — High-quality codebases are crucial for AI agent productivity, as agents excel at pattern recognition.
• Building shared coding guidelines for AI (and people too) — Explicit, pattern-driven coding guidelines are crucial for AI agents to integrate effectively into enterprise codebases.
• Casper: Helping developers work effectively with coding assistants — Sustainable AI coding productivity requires structured human-AI collaboration and embedded engineering discipline, not just smarter AI.
• How (and what) I'm building this week — Interactive, agent-driven learning and development workflows enhance building and deployment efficiency.
• Why I Moved My Agentic Coding Harness from Claude Code to Codex — Daily operational factors like rate limits and reliability outweigh benchmarks for agentic coding harness adoption.
• Writing a Good AGENTS.md — Effective `AGENTS.md` files are concise, human-written, and focus on essential, non-redundant information.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Refactoring.