Task → Agent + Tool + Learning
Work on an Existing Project
On a live codebase the job is orientation first: read the layout, respect project rules, then change a small surface. Agents that live in the editor, plus a review checklist, beat dumping the whole repo into a generic chat.
AI Agents
IDE · Agent & Skills
Cursor
IDE coding agent with tools and MCP; extend behavior with project or global Skills (SKILL.md) for repeatable workflows.
View Agent →IDE · Cascade
Windsurf
AI-native IDE with Cascade agent, terminal tools, and reusable Flows for repeatable dev workflows.
View Agent →IDE · Open source
Continue
Open-source VS Code/JetBrains hub; config.yaml agents and rules for team coding workflows.
View Agent →Review · Security · Quality
Code Review Agent
Review PRs with a consistent checklist
View Agent →AI Tools
Learn
Change a Real Codebase
Cursor as a skill-extensible agent for tickets, tests, and project rules.
Read Guide →Code Review Agent
A review-shaped agent for diffs, risks, and a consistent checklist.
Read Guide →How Agents Use Tools
How agents inspect files, run commands, and stay inside the repo you already have.
Read Guide →Other coding tasks
Build a Website
Create a website or landing page with AI.
View →Build an App
Build web or software applications with AI.
View →Write Code
Generate, edit, and understand code with AI.
View →Fix a Bug
Debug errors and solve coding problems with AI.
View →Learn Coding with AI
Learn programming and development with AI assistance.
View →Frequently asked questions
Can I paste an entire repository into a chat window?
Usually no—and it is a weak way to work anyway. An editor agent that can read files, follow ignore rules, and run commands stays inside the project you already have.
How do project rules help a coding agent?
Rules (and skills) tell the agent your stack, test commands, and what not to touch. That is how you get the same behavior on the next ticket instead of re-explaining the repo.
What if the agent edits the wrong files?
Scope the task: name the directories, require a plan before writes, and review the diff. Start with a read-only “explain this module” pass on an unfamiliar codebase.