IntegrAuth Academy
Identity, security and practical AI — free, from zero. Byte-sized lessons, real diagrams, hands-on labs, zero fluff. Two programs, each with its own exam and certificate.
English is hard? Your browser can translate this page into Tamil, Telugu, Kannada, Malayalam or Hindi. Tech words stay in English so you learn the real names.
How?
- Chrome on phone: ⋮ menu → Translate
- Chrome on computer: right-click → Translate to …
- Edge: the translate icon in the address bar
Set up your coding assistant
Priya wants an AI agent to work in her team's code. Your goal: set up so a bad change is easy to undo.
🧪 Interactive lab — enable JavaScript to play with this one.
git restore.- Git is your undo button. Start clean, on a new branch.
- Run the tests first. Note what already fails.
- Start in the most careful mode. Ask it to explain first.
- Commit good changes. Throw away bad ones.
- Paying per use? Set a spend limit first.
Where it runs
A coding agent reads and edits your files and runs commands. It runs in your terminal (the window where you type commands), your code editor or the cloud. Start on your own computer, where you can watch.
The git safety net
Run git status. It should say "working tree clean": no unsaved changes. If not, commit (save) your own work first, or your edits and the agent's get mixed up. Then make a branch for the agent. A branch is like a photocopy of your notebook. The agent writes on the copy. Your original stays safe.
Next, run the tests, the small programs that check the code. Note which ones already fail. This list is your baseline.
Your first session
Pick your tool's most careful permission mode. Then each edit and command waits for you. It may not be the default. First, ask the agent to explain the project, with no changes. Check it against the README. Then ask for one small change.
Now keep or throw away five changes.
🧪 Interactive lab — enable JavaScript to play with this one.
Go deeper: the three places
A terminal agent is a command you start in your project folder. It uses your files and your logins, as you. You approve its actions as they come. An editor assistant is a panel in your code editor. You accept or reject its edits there.
A cloud agent works on a copy of your code, on the company's computers. It pushes a branch or opens a pull request: a proposed change that waits for review. Nothing joins the main code until a person merges it. You review the finished work, not each step.
Examples, as of October 2026, in no order:
- Terminal: Claude Code, OpenAI Codex CLI, GitHub Copilot CLI, Google's Antigravity CLI, the open-source OpenCode and oh-my-pi (
omp). - Editor: GitHub Copilot, Cursor, the open-source Cline and Kilo Code, the open-source Zed editor.
- Cloud: GitHub's Copilot cloud agent, Codex Cloud, Claude Code on the web.
Google's older Gemini CLI still works for some company accounts.
Go deeper: install and pay
Agents need a terminal and git. Some also need a runtime, like Node.js or Python (Tech basics: terminal, JSON and keys). Installing is usually one command, then a sign-in in your browser. Copy that command only from the official docs. A one-line installer runs a script from the internet as you. A misspelled package name can install someone else's package.
| Sign-in | What stops the bill |
|---|---|
| Subscription | A flat monthly plan. At its usage limit you get paused, not billed more. That changes if you turn on paid extra usage. |
| API key | You pay per token. Set a spend limit. With prepaid credits, keep auto-reload off or low. |
| Free tier | Daily or monthly caps that change. Read the terms. |
| Your own key or a local model | Whatever that provider allows. |
An open-source tool does not decide where your code goes. The model it uses does (safe AI at work).
Agents cost more than chats. Each step sends the whole, growing chat again. On an API key, set a monthly spend limit and an alert first. Keep the key in an environment variable, never in the code or the chat.
Go deeper: undo in detail
git diff shows the changes you have not staged yet. git diff HEAD shows every change since your last commit. git restore . throws away edits to files git already tracks. Did the agent stage them with git add? Then run git restore --staged . first. New files stay. Delete them, then check git status.
Many tools add their own undo, like checkpoints. They help, but they are not git. Some track only the tool's own file edits. Changes made by commands can slip past. So commit each good step in git.
Go deeper: modes
The careful modes are usually "ask every time" or a read-only plan mode. Names differ by tool. Some tools now start in an AI-reviewed mode. A second AI checks each risky action and blocks it, instead of asking you. As of October 2026, Claude Code (version 2.1.283 or later) starts its terminal sessions this way. Company settings can change that. Cursor recommends its AI review as the default. Codex works inside a sandbox and asks before it goes outside. Want to approve each step? Switch to an ask-first mode. Permissions, sandboxes and secrets compares all the modes.
A .gitignore file keeps a file out of git. An agent can still read it and send it to the model provider. Keep real keys out of the project folder and out of the chat. If a key leaks, replace it (secrets management).
Try GitHub Copilot Free. Or try an open-source agent, like Cline, with a local model through Ollama. On a paid plan, try the Claude Code quickstart. Start small: make a branch, run the tests, ask for an explanation.