Author describes moving from JetBrains IDEs to a terminal-based agent workflow using a cascade of AI models: Gemini 3 Flash ($20/mo) for boilerplate generation, GLM-4.7 ($3/mo) as backup, and Claude Code ($100/mo) only for complex logic and polishing. Claims this pipeline covers ~95% of development work (Go/PHP/Python backends) while cutting costs from $250–300/mo (Claude-only) to $123/mo.
Built a Go CLI tool called Gokin (MIT, open source) to orchestrate multi-provider workflows with features like secret redaction, semantic code search, and sandboxed command execution. Argues the shift isn't about replacing IDEs universally but redefining the developer's role from coder to agent architect—especially viable for backend work where visual tooling matters less.
Go-native alternative to Claude Code (TypeScript) that runs locally with multiple affordable backends (Gemini Flash 3, DeepSeek, GLM-4, Ollama). Cuts AI coding costs from ~$100/month to $1–3/month while maintaining speed and security—no Electron overhead, automatic secret redaction, and pure Go performance. Fills a real gap for developers wanting a lightweight, vendor-agnostic coding assistant.
I built Gokin after hitting Claude Code's usage limits and realizing I didn't want to trust third-party CLIs (especially from Chinese AI companies) with direct access to my codebase. It's a Go-based CLI assistant designed to handle the heavy lifting of initial development while keeping sensitive code local.
Key differentiators:
• Cost: GLM-4 backend runs ~$3/month vs Claude Code's ~$100/month. Gemini Flash has a generous free tier.
• Security-first: All code stays local. Automatic secret redaction, sandboxed bash execution, and no telemetry.
• Hybrid workflow: Use Gokin for scaffolding/bulk operations → Claude Code for polish/architectural refinement.
• Rich tooling: Semantic code search (embeddings), multi-agent planning (Beam/MCTS/A*), MCP server integration, Git workflow, and undo/redo for file operations.
• Provider-agnostic: Works with Gemini 3 Flash/Pro (free tier) or GLM-4.7 (~$3/mo).
Tradeoffs: Chinese models (GLM) still lag behind frontier models for complex reasoning—but for scaffolding, refactoring, and repetitive tasks, they're surprisingly capable. Gemini Flash closes much of the gap with a free tier.
Built entirely in Go with a TUI (Bubble Tea). MIT licensed.
Built a Go CLI tool called Gokin (MIT, open source) to orchestrate multi-provider workflows with features like secret redaction, semantic code search, and sandboxed command execution. Argues the shift isn't about replacing IDEs universally but redefining the developer's role from coder to agent architect—especially viable for backend work where visual tooling matters less.