AI coding assistants

AI coding assistants have moved beyond autocomplete. This is a brief overview of what to look for and how they differ from AI features inside code editors.

AI assistants vs AI in editors

Modern code editors like VS Code, Zed, and Cursor offer built-in AI features — inline autocomplete, a chat panel, and commands to rewrite or explain a selection — but AI coding assistants are a distinct category that operates at a higher level of autonomy, enabling them to read your entire codebase, edit multiple files, run shell commands, and execute multi-step tasks from a single instruction. These editor features excel at small, focused tasks like completing a line or generating a short function, while standalone assistants handle more complex workflows.

Some tools in this section are standalone CLIs or web agents. Others sit inside an editor. The distinction is not always sharp — Copilot and Codeium started as autocomplete tools and have since added agentic capabilities, while Claude Code is primarily an AI agent.

What to look for

Model quality is the most important factor when selecting an AI coding assistant, since the underlying language model determines how well the tool understands code, handles edge cases, and avoids mistakes, and since quality varies significantly across providers and improves regularly, checking recent benchmarks before committing to a tool is essential.

Agentic capability describes how much the tool can do on its own. A basic assistant makes suggestions; a capable agent can edit files, run tests, and iterate on a task until it is done. More autonomy is not always better — it depends on your workflow and how much you trust the output.

Editor integration varies. Some assistants live entirely in the terminal; others integrate deeply into your editor with diff views, accept/reject controls, and context pickers. Consider how much you want to stay in your editor vs working in a terminal.

Model selection matters if you have preferences about which AI provider processes your code. Some tools are locked to a single provider; others let you choose.

Privacy and data handling is a significant concern when working with sensitive codebases. All cloud-based assistants send some code context to external servers. Check each tool's data handling policy before using it on proprietary or regulated code. Local model support is an option for stricter environments, though quality is typically lower.

Pricing ranges from free tiers with limited usage to per-seat subscriptions and API-based pay-as-you-go plans.

Integration with fortrabbit

An assistant reads the repository on its own, but the hosting account behind it is a blank spot. It cannot see which environment runs which PHP version, or why last night's deployment failed. Three pieces close that gap, and they stack.

MCP server — a live connection to the account. The agent lists apps and environments, reads the build output of a failed deployment, clones an environment or sets a variable. Authorization runs in the browser, so no token is pasted into a config file. Setup is one command per client:

MCP setup instructions

https://www.fortrabbit.com/mcp

Agent skills — the workflow knowledge MCP does not carry: how a deployment is done, how a database is synced, which command to run over SSH. Installed once, they stop an agent from guessing at platform conventions.

CLI and public API — the fallback for anything the tool set does not expose. An assistant that can run shell commands can run frbit, and the API covers the rest.

Support differs per assistant. The pages below note what each one connects to.