The AI Coding Tool Landscape

This content was generated with the assistance of AI and should be verified against the official documentation of each tool and framework before being relied on in production.

This section’s bibliography lists the reference material consulted while preparing these pages.

This page orients the rest of the sub-section: it names the categories of AI coding tool in use today, places eight current tools on a dated comparison table, and gives a decision table for picking a category by task. The facts below were checked against each tool’s own documentation as of 2026-09-28. Two tools anchor the rest of this sub-section and get their own getting-started page: GitHub Copilot (VS Code 1.139, Copilot Chat, and the Copilot CLI, all re-verified on that date) and Claude Code (the current CLI, checked with claude --version against its own changelog on that date). The running example throughout this sub-section is DocsAssistant API, a small Spring Boot + Python sample repo (src/main/java/com/example/docsassistant/DocsController.java, tests under src/test/java/…, an ingestion script under ingest/), the same scenario used in LLM Foundations & Prompting.

Tool categories

Every tool below fits at least one of six categories. A single product can span several: GitHub Copilot, for example, ships a completion engine, a chat panel, an agent mode and a cloud agent all at once.

Completion: Inline "ghost text" that continues the line or block you are typing, plus next-edit suggestions that predict a related edit elsewhere in the file. It needs no prompt beyond the code and comment already there. Example: typing // return all pages whose title contains the given keyword above an empty method in DocsController.java and accepting the suggested body with Tab. See GitHub Copilot — Completions.
IDE chat: A conversational panel inside the editor that answers questions and can edit the file you are looking at, but only when you ask; it does not act on its own. Example: asking "Explain what this method does and list its edge cases" in the chat panel with DocsController.java open. See VS Code — Use chat in VS Code.
IDE agent: A mode where the assistant plans a multi-file change, edits, runs commands (build, test, lint) and reports back, asking for approval at configured points. Example: "Add a GET /api/pages/{id} endpoint with a test" run in agent mode inside the IDE. See VS Code — Build with agents in VS Code.
Terminal agent: The same agentic loop, run from the command line instead of an IDE, so it composes with shell pipes, scripts and CI. Example: claude "write tests for DocsController and run them". See Claude Code — Overview.
Cloud / background agent: An agent that runs unattended in its own remote sandboxed environment — typically started from an issue, a PR comment or a chat mention — and reports back with a pull request once it is done. Example: assigning a GitHub issue to Copilot and letting it open a draft PR while you do something else. See GitHub Copilot — About Copilot cloud agent.
App builder: A prompt- or screenshot-driven tool that generates a runnable UI or small app from scratch, optimized for a fast prototype rather than for editing an existing large codebase. Example: "Build a page that lists DocsAssistant’s indexed pages with a search box" typed into a fresh v0 project. See v0 by Vercel — What is v0?. This category is covered in depth on Prototyping with AI.

The autonomy ladder

The categories above sit on a single ladder of increasing autonomy: each rung keeps everything the rung below it can do, and adds the ability to act with less turn-by-turn supervision.

flowchart LR A["Completion
finishes what you type"] --> B["Chat
answers when asked"] B --> C["Agent
plans, edits, runs, verifies"] C --> D["Background agent
runs unattended, reports back"]

App builders sit alongside this ladder rather than on it: they are agentic in the same sense as an IDE or terminal agent, but scoped to generating a new app rather than working inside an existing one. The ladder itself — and the "70 % problem" of the extra effort needed to take an agent’s output the rest of the way — is introduced on AI-Assisted Development and covered in more depth on Agent Mode.

Comparison of today’s AI coding tools (as of 2026-09-28)

Facts below were re-checked against each linked page on that date. Prices, release cadences and model lists change too often to keep current here; see each tool’s own pricing and supported-models pages, linked from its docs home, instead.

Tool	Category	Where it runs	Agentic?	Background / cloud agent?	MCP support?	Official docs
GitHub Copilot	Completion, IDE chat, IDE agent, cloud agent, terminal agent	IDE extensions (VS Code, JetBrains, Visual Studio, Xcode), github.com, a terminal (Copilot CLI), and a standalone GitHub Copilot desktop app	Yes — agent mode in supported IDEs	Yes — the Copilot cloud agent researches, plans and edits in its own ephemeral GitHub Actions environment	Yes	docs.github.com/en/copilot
Claude Code	Terminal agent, IDE agent, cloud agent	A terminal CLI, IDE extensions (VS Code, JetBrains), a desktop app, and the web at claude.ai/code	Yes — an agentic loop that reads, plans, edits, runs and verifies	Yes — cloud sessions on the web and mobile, a GitHub Actions integration, and scheduled routines	Yes	code.claude.com/docs/en/overview
Cursor	IDE agent, terminal agent, cloud agent	A standalone AI-first editor, a CLI (Cursor CLI), plus cursor.com/agents on the web	Yes — documented as "a coding agent for building ambitious software"	Yes — Cloud Agents (the current name for what were called background agents), one dedicated VM per task	Yes	cursor.com/docs
Cline	IDE agent, terminal agent	An extension for VS Code, JetBrains, Cursor, Windsurf, Zed and Neovim; a CLI; a desktop app	Yes — an autonomous agent that asks approval at each step by default	Partial — the desktop app adds parallel sessions and cron-style scheduled tasks; the docs do not describe a fully hosted cloud agent comparable to the others in this table	Yes	docs.cline.bot
OpenAI Codex	IDE agent, terminal agent, cloud agent	The ChatGPT desktop app, an IDE extension, a CLI, and Codex cloud on the web	Yes	Yes — Codex cloud runs isolated, parallel and schedulable cloud tasks	Yes	developers.openai.com/codex
Google Jules	Cloud / background agent	Cloud — Jules clones the repository into a virtual machine and works asynchronously; started from the web or from a terminal with the Jules Tools CLI	Yes	Yes — this is the whole product: submit a task, keep working elsewhere, get a browser notification when done	Yes — MCP server connections, added on 2026-02-02	jules.google/docs
Gemini Code Assist	Completion, IDE chat, IDE agent	IDE extensions (VS Code, JetBrains, Android Studio), Cloud Shell Editor, Cloud Workstations, and the separate Gemini CLI in a terminal	Yes — agent mode runs multi-step tasks and can call MCP servers	Not documented as a standalone background/cloud coding agent comparable to the others in this table	Yes — agent mode can call MCP servers	docs.cloud.google.com/gemini/docs/codeassist
Amazon Q Developer	Completion, IDE chat	IDE plugins only (VS Code, JetBrains, Visual Studio, Eclipse); the former Q Developer CLI is now Kiro CLI	Partial — the IDE plugins are chat- and completion-focused	Not documented as a background/cloud coding agent	Yes — the IDE plugins can call MCP servers (see MCP configuration in the IDE)	docs.aws.amazon.com/amazonq

Tool

Category

Where it runs

Agentic?

Background / cloud agent?

MCP support?

Official docs

GitHub Copilot

Completion, IDE chat, IDE agent, cloud agent, terminal agent

IDE extensions (VS Code, JetBrains, Visual Studio, Xcode), github.com, a terminal (Copilot CLI), and a standalone GitHub Copilot desktop app

Yes — agent mode in supported IDEs

Yes — the Copilot cloud agent researches, plans and edits in its own ephemeral GitHub Actions environment

Yes

docs.github.com/en/copilot

Claude Code

Terminal agent, IDE agent, cloud agent

A terminal CLI, IDE extensions (VS Code, JetBrains), a desktop app, and the web at claude.ai/code

Yes — an agentic loop that reads, plans, edits, runs and verifies

Yes — cloud sessions on the web and mobile, a GitHub Actions integration, and scheduled routines

Yes

code.claude.com/docs/en/overview

Cursor

IDE agent, terminal agent, cloud agent

A standalone AI-first editor, a CLI (Cursor CLI), plus cursor.com/agents on the web

Yes — documented as "a coding agent for building ambitious software"

Yes — Cloud Agents (the current name for what were called background agents), one dedicated VM per task

Yes

cursor.com/docs

Cline

IDE agent, terminal agent

An extension for VS Code, JetBrains, Cursor, Windsurf, Zed and Neovim; a CLI; a desktop app

Yes — an autonomous agent that asks approval at each step by default

Partial — the desktop app adds parallel sessions and cron-style scheduled tasks; the docs do not describe a fully hosted cloud agent comparable to the others in this table

Yes

docs.cline.bot

OpenAI Codex

IDE agent, terminal agent, cloud agent

The ChatGPT desktop app, an IDE extension, a CLI, and Codex cloud on the web

Yes

Yes — Codex cloud runs isolated, parallel and schedulable cloud tasks

Yes

developers.openai.com/codex

Google Jules

Cloud / background agent

Cloud — Jules clones the repository into a virtual machine and works asynchronously; started from the web or from a terminal with the Jules Tools CLI

Yes

Yes — this is the whole product: submit a task, keep working elsewhere, get a browser notification when done

Yes — MCP server connections, added on 2026-02-02

jules.google/docs

Gemini Code Assist

Completion, IDE chat, IDE agent

IDE extensions (VS Code, JetBrains, Android Studio), Cloud Shell Editor, Cloud Workstations, and the separate Gemini CLI in a terminal

Yes — agent mode runs multi-step tasks and can call MCP servers

Not documented as a standalone background/cloud coding agent comparable to the others in this table

Yes — agent mode can call MCP servers

docs.cloud.google.com/gemini/docs/codeassist

Amazon Q Developer

Completion, IDE chat

IDE plugins only (VS Code, JetBrains, Visual Studio, Eclipse); the former Q Developer CLI is now Kiro CLI

Partial — the IDE plugins are chat- and completion-focused

Not documented as a background/cloud coding agent

Yes — the IDE plugins can call MCP servers (see MCP configuration in the IDE)

docs.aws.amazon.com/amazonq

Two currency notes, checked on the same date: the Amazon Q Developer CLI was rebranded as Kiro CLI on 2025-11-17 (see AWS — Upgrade to Kiro and Kiro — Upgrading from Q CLI), so the terminal agent with MCP support now ships under the Kiro name; and AWS states that the remaining Amazon Q Developer IDE plugins reach end of support on 2027-04-30 (see AWS — Amazon Q Developer end-of-support announcement). And "MCP support" in the table means each tool can act as an MCP client, connecting to external MCP servers; none of the columns above are about building an MCP server, which belongs to the Model Context Protocol (MCP) sub-section.

For the three tools this account’s own AI-assisted development skills are built on — Claude Code, GitHub Copilot and OpenAI Codex — see Claude Code vs. GitHub Copilot vs. OpenAI Codex for a feature-by-feature comparison rather than repeating it here.

When to use which

Task	Best-fit category	Why
A one-line edit while typing (a getter, a loop bound, an obvious continuation)	Completion	Ghost text finishes the line in place; there is no round trip to a chat panel to interrupt typing flow.
"Explain this" or "what does this parameter do" about code already open	IDE chat	The answer sits next to the code, using open tabs and the current selection as context, with no file changes.
"Add an endpoint and its tests" spanning a controller, a service and a test file	IDE agent	The agent plans across files, runs the build and test commands itself, and presents one reviewable diff.
A scripted, repeatable task: a formatting pass, a bulk rename, a CI step	Terminal agent	Composable with shell pipes, flags and exit codes, so it fits a script or a CI job rather than a chat UI.
"Pick up this issue and open a PR" while you are away from the keyboard	Cloud / background agent	Runs unattended in its own sandboxed environment and comes back with a pull request to review later.
"Turn this idea or screenshot into a working UI, fast"	App builder	Optimized for a working prototype from a short prompt or an image, not for editing a large existing codebase.

Task

Best-fit category

Why

A one-line edit while typing (a getter, a loop bound, an obvious continuation)

Completion

Ghost text finishes the line in place; there is no round trip to a chat panel to interrupt typing flow.

"Explain this" or "what does this parameter do" about code already open

IDE chat

The answer sits next to the code, using open tabs and the current selection as context, with no file changes.

"Add an endpoint and its tests" spanning a controller, a service and a test file

IDE agent

The agent plans across files, runs the build and test commands itself, and presents one reviewable diff.

A scripted, repeatable task: a formatting pass, a bulk rename, a CI step

Terminal agent

Composable with shell pipes, flags and exit codes, so it fits a script or a CI job rather than a chat UI.

"Pick up this issue and open a PR" while you are away from the keyboard

Cloud / background agent

Runs unattended in its own sandboxed environment and comes back with a pull request to review later.

"Turn this idea or screenshot into a working UI, fast"

App builder

Optimized for a working prototype from a short prompt or an image, not for editing a large existing codebase.

One example per category, in its native form

Completion — a comment that grows into a method

// return all pages whose title contains the given keyword, case-insensitively

Typed above an empty method in DocsController.java; Copilot’s ghost text completes the method body beneath it. See GitHub Copilot — Completions.

IDE chat — a question about open code

Explain what DocsController.getPageById does and list its edge cases.

Asked in the VS Code Chat view with DocsController.java open. See VS Code — Use chat in VS Code.

IDE agent — a multi-file change

Add a GET /api/pages/{id} endpoint to DocsController, backed by PageService, with a passing test.

Run in VS Code’s agent mode. See VS Code — Build with agents in VS Code.

Terminal agent — the same task from a shell

$ claude "add a GET /api/pages/{id} endpoint to DocsController, backed by PageService, with a passing test"

See Claude Code — Overview.

Cloud / background agent — delegating a whole issue

@claude please pick up issue #42 (add pagination to GET /api/pages) and open a PR.

Posted as a GitHub issue or PR comment once the Claude GitHub App is installed. See Claude Code — GitHub Actions. The Copilot cloud agent equivalent is assigning the issue directly to Copilot; see GitHub Copilot — About Copilot cloud agent.

App builder — a prompt that generates a new UI

Build a page that lists DocsAssistant's indexed pages with a search box, using a simple card layout.

Typed into a new v0 project. See v0 by Vercel — What is v0?.