Choice map

AI tools and models: IDEs, CLIs and LLMs

intermediate 18 min read

In 30 seconds

Understand the difference between tools, products, and models so you can choose IDEs, CLIs, cloud agents, and LLMs with intent.

Two different choices

When someone asks “which AI should I use for coding?”, they are usually mixing two decisions:

  1. Tool: where you work. It can be an IDE, a CLI, a cloud app, an editor with a plugin, or a prompt-to-app platform.

  2. Model: the brain that generates the answer. It might be from the GPT, Claude, Gemini, Kimi, Llama, Gemma, DeepSeek, GLM, or another family.

These layers combine, but they are not the same thing. Cursor and Claude Code are tools. GPT/Codex-line, Claude Sonnet/Opus, Gemini, and Kimi models are example brains. ChatGPT, Claude, Gemini, and Kimi are products that can embed different models underneath.

The simple rule: product and tool are where you interact. Model is what generates or decides the next step.

Three work surfaces

Market note: the names below are a July 2026 snapshot. Categories last longer than brands.

Every week a new tool promises to revolutionize how you program. Instead of listing everything that exists, think in three surfaces:

  1. AI-native IDEs: code editors with AI built into the editing experience. Examples: Cursor, Windsurf, VS Code with GitHub Copilot, Google Antigravity.

  2. AI CLIs: command-line tools that read files, run commands, and make changes directly in the terminal. Examples: Claude Code, Codex CLI, OpenCode, Kimi Code, GitHub Copilot CLI.

  3. Cloud apps and agents: you send a well-described task, the tool works in an isolated environment, researches the repo, proposes a plan, works on a branch, and returns a diff or pull request. Examples: Codex (cloud), Google Jules, GitHub Copilot coding agent, Devin.

The boundary is getting blurry. The same vendor shows up in the terminal, the editor, and the cloud. Categories still help, but the main question is not “which tool is best?”. The question is: which surface removes the most friction for this task?

In cloud agents, an emerging pattern is Research → Plan → Code → Review: research the repo, propose a plan, execute on a branch, and only open a PR after human review. No vendor owns that flow, but more tools converge on it.

What changes across models

You do not need to become an AI researcher to work well with LLMs. A few concepts still change decision quality:

Context

Models process text in tokens and each model has a context window. In 2026, windows range from tens of thousands of tokens into the million range, but an average project can be much larger than that. Context helps, but it does not remove the need to choose relevant files, examples, and constraints.

Reasoning, speed, and cost

Larger models often reason better on long tasks, and they also tend to be slower and more expensive. Using the strongest model for everything can be waste. An exploratory chat, autocomplete, and an architecture plan do not need the same power.

Tool use

Some models are tuned to operate well with tools: read files, call APIs, run commands, observe the result, and decide the next step. That matters a lot for coding agents, because quality depends on more than generated text. It also depends on following a task across many steps.

Real limits

Every model hallucinates, has a knowledge cutoff, and can generate code that looks right but does not solve the problem. Benchmarks are signals, not oracles. A leaderboard measures standardized tasks; your project has context, history, constraints, and its own debt.

How to choose without the hype

Start from the task, not the ranking:

  • fast exploration and autocomplete: IDE surface + light or mid model;
  • multi-file refactor and deep investigation: agent CLI or IDE with a strong agent;
  • long task without you on the shoulder: cloud agent with a plan and a reviewable PR;
  • privacy, cost, or control: open-weight models or BYOK on an open-source CLI.

Tools and models do not replace method. Without clear scope, validation, and human review, the “best” surface only creates debt faster.

Next lesson Professional flow

SDD and Harness Design: professional AI-native development

Keep going

Community

Ask, answer, get unstuck

Use this space to ask questions about the lesson, share examples, and help other people understand the topic.