The AI Coding Tool Landscape

Before you can direct an AI assistant in an interview, it helps to know the territory. The AI coding tool landscape has grown quickly, and where each tool sits shapes what it's good at — and what it isn't. If you understand the categories, you'll understand why the interview environment feels the way it does, and why your daily driver behaves differently.

The four categories

Terminal agents

These run in your terminal, read your whole codebase, edit files, run commands, and verify their own work. They're the closest thing to a senior pair partner: describe a feature, get a plan, approve it, and watch it execute across multiple files.

• Claude Code (Anthropic) — the reference terminal agent. Reads the codebase, edits files, runs commands, and works across surfaces: terminal, IDE extensions, desktop app, and web. Supports plan mode, subagents, hooks, skills, and MCP connections.
• Codex CLI (OpenAI) — a lightweight coding agent that runs locally, sign-in-with-ChatGPT, with sandboxed approval modes and IDE extensions.
• OpenCode — an open source agent (160K+ GitHub stars) that runs in the terminal, as a desktop app, or as an IDE extension, with support for 75+ model providers.

IDE-native

These live inside your editor, offering inline diffs, chat panes, and context from your open files. Cursor and Windsurf are purpose-built AI editors; GitHub Copilot works inside VS Code and JetBrains. They're great for day-to-day work, and several companies suggest Cursor specifically for open-ended interviews.

Chat-first

Web apps like claude.ai and ChatGPT let you paste code and get answers, but they don't have direct access to your filesystem or terminal. They're useful for quick questions, but as an interview tool they leave the "read the codebase, run the tests" part to you.

In-browser interview platforms

CoderPad and HackerRank now ship AI assistants inside their environments. These are the tools you'll actually use in a structured interview. They have an editor, a chat panel, and (usually) your starter code — but they typically lack the system prompts and project context that make terminal agents feel smart. That's why candidates consistently report them feeling "nerfed" compared to daily drivers.

The capability axes that matter in interviews

When you're choosing a daily driver or evaluating an interview environment, these axes matter most:

Axis	Why it matters
Plan mode	A mode that proposes changes without making them. This is your rehearsal tool for "plan first, then execute."
Codebase context	Can the tool read your whole repo, or just what you paste? This decides whether prompts referencing class names actually land.
Test-running	Can the agent run tests and iterate on failures, or do you relay errors manually?
Undo / checkpoints	When the agent goes off the rails, can you roll back cleanly?
Diff review	Can you see exactly what changed before accepting it?
MCP / tools	Connections to external systems — useful, but rarely required in an interview.

Picking a daily driver

The interviewers we've talked to don't care which tool you use — they care whether your workflow is real. Pick one terminal agent (or IDE) and get fluent: know how to enter plan mode, how to reference files, how to run tests through it, and how to undo. Then rehearse with the interview tool, which will be less capable than your driver, so the gap doesn't surprise you on the day.

The next three articles go deep on the three major terminal agents — Claude Code, Codex, and OpenCode. Each covers setup, the core workflow, and the habits that transfer to an interview.