AI Coding Tools

Choose the right AI coding tool and learn how to write effective prompts for each one. Every tool has different prompt patterns, configuration files, and capabilities. This section covers them all.

Quick Comparison

Which Tool Is Right for You?

Claude Code

Best for complex multi-file refactors, autonomous coding sessions, and extended thinking. Requires a Claude subscription or Anthropic Console account. Available in the terminal, IDE extensions, desktop app, and web. Full guide → Patterns →

OpenCode

Best for provider flexibility and cost optimization. Supports 75+ LLM providers, switch mid-session, custom skills, Code Mode, and background subagents. Free, Console, and Go pricing tiers. Full guide → Patterns →

Oh My Pi (OMP)

Best for developers who want Pi's terminal-first model with more built-ins: provider logins, resumable sessions, and a guided setup path. Install and setup →

Gemini CLI

Best for Google ecosystem users and large codebases. 1M+ context window, sandbox modes, Google Drive/Calendar/Gmail extensions. The consumer free tier was replaced by Antigravity CLI in June 2026. Full guide →

Cursor

Best for IDE-first developers. Tab completions, inline editing, agent mode, composer. Best .cursorrules ecosystem. $20/mo. Full guide →

GitHub Copilot

Best for Microsoft ecosystem and GitHub workflow integration. Workspace context, agent mode, PR/issue integration. Free/$10/mo. Full guide →

Tool-Specific Prompt Engineering

Each tool has unique prompt patterns and configuration systems. Master them:

• Claude Code — CLAUDE.md patterns, extended thinking, hook scripts
• OpenCode — AGENTS.md conventions, /slash commands, provider switching
• Gemini CLI — Ralph loop, sandboxing, system instruction files
• Cursor — Agent mode vs inline, .cursorrules, composer prompts
• GitHub Copilot — copilot-instructions.md, workspace context, agent mode

Rules and instruction files are only half of these tools' context: the reusable procedures each agent loads on demand — review passes, test loops, cleanup steps — are packaged as skills, and the site keeps them in one curated skills collection.

Decision Guide

Decision Models

Not every AI call is a coding call. Jev is TypeSafe's System One decision model: send typed state and questions, get probabilities instead of prose. Use it to classify, route, gate, verify, and score — the decisions an agent currently pays a chat model to make. See the Jev guide for setup, prompting rules, and integration patterns.

And not every agent finding needs a vector stack. Evoke is an Apache-2.0 Postgres extension for learned-sparse retrieval: keyword and meaning evidence share one inverted index and one score, with a 30M-parameter model on CPUs inside the database. Exact codes still lead while "the words I didn't write" get found. See the Evoke guide for install, use cases, and the exact-vs-paraphrase pattern.

Other Notable Tools

• Windsurf — Cascade AI flow, multi-file editing, rules system
• Continue.dev — Open-source, local models, custom slash commands
• Aider — Git-native, map-refine architecture, polyglot
• Cline — VS Code plugin, autonomous coding, browser automation
• Codex CLI — OpenAI's CLI agent, reasoning strategies, sandbox
• Roo Code — Cline fork, specialized modes, VS Code plugin
• Tabnine — Enterprise AI assistant, code privacy, on-premise

Not Sure Where to Start?

See the Tool Comparison guide for a detailed head-to-head analysis across pricing, context windows, prompt patterns, MCP support, and more. Pick the right tool for your stack and workflow.

Sources

• Claude Code overview
• Claude Code model configuration
• Gemini CLI repository
• Cursor MCP documentation
• GitHub Copilot MCP documentation