This guide breaks down the four major workflow types you’ll encounter—IDE agents, terminal agents, pull‑request agents, and cloud agents—and explains how each one fits into real development work.

What Makes a Coding Tool “Agentic”?

Unlike traditional chatbots that answer a single prompt and stop, coding agents work in a loop:

1. Read relevant files or context

2. Reason about the steps needed

3. Act by editing code or running commands

4. Evaluate the outcome and continue if needed

This cycle repeats until the task is complete or the agent hands control back to you.
Where this loop runs—your editor, your terminal, your CI system, or the cloud—defines the workflow type.

The Four Workflow Types

1. IDE Agents

IDE agents live directly inside your editor. They provide real‑time suggestions, inline edits, and visual diffs. You stay in full control, accepting or rejecting changes as you go.

Two main styles exist:

• AI‑native IDEs (Cursor, Windsurf, Kiro)
  Built around agent workflows from the ground up. Some support spec‑driven development where you describe the goal and let the agent execute.

• IDE extensions (GitHub Copilot, Claude Code for VS Code, Gemini Code Assist)
  Add agentic features to your existing editor. These tend to focus on targeted edits and refactoring.

Because many IDE agents rely on cloud inference, teams with strict privacy rules often require local‑only models or approved tools.

2. Terminal Agents

Terminal agents run inside your shell. You describe a task, and the agent:

• Reads your project

• Proposes edits

• Runs commands

• Asks for approval at each step

This workflow shines when you’re navigating large codebases or coordinating changes across multiple files. It’s also great for onboarding into unfamiliar projects.

Popular examples include Claude Code, Aider, Gemini CLI, OpenCode, and Codex CLI.

Terminal agents integrate naturally with existing developer workflows—piping logs, chaining commands, or embedding them in scripts. Some support local models for teams that cannot send code to external servers.

3. Pull‑Request (PR) Agents

PR agents operate asynchronously. Instead of interacting with them live, you open or update a pull request, and the agent reviews it in the background.

They typically:

• Scan diffs

• Flag potential bugs

• Suggest improvements

• Leave comments for human review

This workflow fits team environments where code review is mandatory. Tools like CodeRabbit and GitHub Copilot Code Review support this model.

Because PR agents operate on shared repositories, organizations often manage their permissions centrally.

4. Cloud Agents

Cloud agents offer the highest autonomy. You describe a task, and the agent works in a managed environment—often spinning up compute, running code, and returning results as a branch, PR, or prototype.

They’re ideal for:

• Prototyping

• Long‑running tasks

• Work you don’t want to supervise step‑by‑step

Examples include Devin, Claude Code on the web, Codex web, and Cursor’s Cloud Agents.

Execution models vary: some run on vendor infrastructure, others can run on machines you control. Because they operate remotely, privacy and compliance considerations are essential.

When Categories Overlap

Many modern tools span multiple workflows. For example:

• Claude Code works in IDEs, terminals, PR reviews, and the cloud.

• Cursor covers IDE, CLI, cloud execution, and automated PR review.

• GitHub Copilot supports IDE editing, terminal workflows, PR review, and cloud agents.

The key insight:
The workflow describes how you interact with the tool—not the tool itself.

Common Pitfalls to Avoid

• Using one agent for everything
  Each workflow excels at different tasks. Match the agent to the job.

• Ignoring privacy constraints
  Cloud‑backed agents may send code to external servers. Some teams require local models or self‑hosted environments.

• Over‑automating without review
  AI‑generated code can contain subtle bugs. Human review is essential, especially with highly autonomous agents.

Treat agents as collaborators, not replacements.

Conclusion

Agentic coding is reshaping software development. Understanding the four workflow types—IDE, terminal, PR, and cloud—helps you choose the right interaction model for each task.

• Use IDE agents for interactive editing.

• Use terminal agents for complex, multi‑file changes.

• Use PR agents for automated review.

• Use cloud agents for background tasks and prototypes.

As tools evolve, these categories will continue to blend, but the workflows remain a reliable mental model for choosing the right approach.