Stop asking AI for answers. Start handing it the job.your inbox.the bug.the busywork.the mission.

The AI you already know

Let's start on familiar ground. Grok chat — like any AI chatbot — is a conversation. And a conversation, no matter how smart, has one hard limit.

Think of chat AI as a brilliant expert locked in a room with a mail slot. You slide a note in. A great note slides back out. The expert has read half the internet, reasons well, writes fast, and never gets tired of your questions.

But the expert never leaves the room. It can't open your files, click anything, run anything, or check whether its advice actually worked. Its only tool is words. Everything it "does" happens inside the chat window — and the moment the answer needs to touch the real world, that job transfers to you.

What chat is genuinely great at

• Answers and explanations — "What's compound interest?" "Why is my sourdough flat?"
• Drafts and rewrites — emails, essays, captions, code snippets you'll paste somewhere.
• Thinking out loud — brainstorming, pros/cons, "explain this like I'm new."

None of this goes away in the agentic world. Chat stays the right tool for a huge share of what you'll do. The point of this course isn't "chat is obsolete" — it's that chat is one mode, and until recently it was the only mode you had.

What "agentic" actually means

An agent is AI with three things a chatbot doesn't have: a goal it owns, tools it can use, and a loop it runs until the goal is done.

Unlock the expert's door and hand them a toolbox — the ability to read and edit files, run programs, browse the web, use apps. Now instead of asking a question, you assign a task. The AI breaks it into steps, does a step, looks at what happened, and adjusts. Over and over, until it's finished or needs your input.

That "looks at what happened" part is the magic. Step through the loop below with a real example.

Why this is fundamentally more powerful

Not "somewhat better." A different category. Here's the honest breakdown:

1. It acts instead of advising. Chat's finished product is instructions. An agent's finished product is the finished product. The 50 photos are renamed. The bug is fixed. The report exists.
2. It checks its work against reality. A chatbot can't know if its advice worked. An agent runs the thing, reads the error message, and fixes its own mistake — errors become feedback instead of dead ends. This self-correction is the single biggest upgrade.
3. It carries long chains of steps. A real task might be 40 small steps. With chat, you shepherd every one — you're the project manager. An agent holds the whole chain itself and comes back when it's done or stuck.
4. It works with your real, current stuff. Chat only knows what you paste into the window. An agent reads the actual files, the actual inbox, the actual codebase — no copy-paste bottleneck.
5. It scales past your attention. You can only type in one chat at a time. Agents can run while you're away, on schedules, and even in parallel. Your time stops being the ceiling.

Meet Grok Build — the agent on your computer

Grok Build is xAI's coding agent. It runs in your terminal, inside your project folder, with its hands on your actual files.

First: what's a terminal? (30 seconds)

The terminal is the text-only window where you control your computer by typing commands instead of clicking icons. It looks intimidating — black screen, blinking cursor — but it's just another way to talk to your machine, and it's where programmers live. Grok Build moved in there for a reason: the terminal is where you can do anything — create files, run programs, install software. An agent living there inherits all of that reach.

Here's the workflow: you open a terminal, go to a project folder, and type one word — grok. A full-screen session opens. Then you talk to it in plain English, and it works on the code around it. Watch:

Read that back and you'll spot every part of Module 2's loop: it ran the tests (act), read the failure (check), found the cause in the code, asked permission before editing, then re-ran the tests to prove the fix. Chat could only have guessed at that bug. Build watched it fail and watched it pass.

The spec sheet, in plain English

So how is this different from asking Grok chat about code?

• Chat sees what you paste. Build reads the actual project — all of it — and keeps its understanding as it works.
• Chat suggests an edit. Build makes the edit, in the real file, and shows you the diff.
• Chat can't run your tests. Build runs them before and after, so "fixed" means verified fixed, not "probably fixed."
• Chat forgets your project tomorrow. Build works where your project lives, so every session starts in context.

Meet Grok Bot — the crew in the cloud

Grok Bot flips the setup. Instead of an agent visiting your computer, xAI gives your agents their own computer — a dedicated machine in the cloud where they work around the clock, whether or not you're online.

Launched in August 2026, Grok Bot is less "a chatbot" and more "a small staff." You create named bots — think of them as teammates with job descriptions — and message them the way you'd message a coworker. Behind each friendly chat window is a real computer with a browser, a file system, and a terminal, shared by your whole crew.

The three ideas that make it click

1. One shared cloud computer. All your bots work on the same dedicated machine, with the same browser, files, and connected accounts. Connect Gmail or Slack once and the whole crew can use it. When there's no official connector, a bot shows you the login screen so you sign in yourself — bots don't collect your passwords.
2. Routines. A routine is a standing instruction: run on a schedule ("every weekday at 8am") or on an event ("when a new email arrives," "when someone pushes a commit"). This is the moment AI stops waiting for your message.
3. Teach by showing. Record yourself doing a task once — stepping through your expense tool, say — and Grok Bot turns the recording into a repeatable skill the bot can run and refine on its own.

How is this different from Grok chat?

• Chat starts when you type and stops when you close the tab. A bot has a standing job. Close the app; it keeps working.
• Chat is one anonymous window. Bots are named, persistent teammates that remember their role and your preferences.
• Chat can describe an email you paste in. A bot is in your inbox (with your permission), triaging the real thing.
• Chat answers. A bot delivers — drafts sent for approval, reports posted, trackers updated.

Build vs Bot: same brain, different body

Here's the graduate-level insight: Grok Build and Grok Bot run on the same Grok models and the same agent loop you learned in Module 2. What differs is where the agent lives, what it can reach, and how it relates to your time. Those three differences change everything about when you'd use each.

1 · Where it lives decides what it touches

Build lives on your machine. Its world is the project folder you start it in — your real files, your installed tools, your code. Nothing needs to leave your computer, which is exactly what you want when the work is your computer: a codebase, a folder of files, a local project.

Bot lives on xAI's cloud. Its world is a dedicated online computer plus the accounts you've connected — email, Slack, calendar, the web. That's exactly what you want when the work lives in your accounts and apps rather than on your disk.

2 · Session vs standing job

Build is a session. You sit down, open the terminal, work with it — often watching, steering, approving — and when you close it, it's done. It's a power tool: enormous leverage while your hands are near it.

Bot is a standing job. You hire it once, give it routines, and it runs whether you're present or asleep. It's less a power tool than an employee — the value shows up on a schedule, not per-session.

3 · Driver's seat vs delegation

With Build, you're in the loop by default: it proposes an edit, you approve; it shows a diff, you read it. With Bot, autonomy is the point — you review the outputs (a draft held for approval, a posted report) more than the steps. That's also why Bot's guardrails center on account access and approval queues, while Build's center on file edits and command permissions.

Prove it — the scenario game

Six real-life situations. Pick the right tool: Chat Build Bot

Scored all six? Then you've internalized the real lesson: it's never "which AI is best" — it's where does the work live, and when should it happen.

Your first mission

You now understand agentic AI better than most people on the internet. Here's how to turn that into hands-on experience this week — safely.

The four golden rules for new agent users

1. Start read-only. Your first tasks should look, not touch: "explain this codebase," "summarize my unread email." Zero risk, instant feel for how agents think.
2. Review before you approve. When an agent asks "apply this change?", actually read it. You're training your own judgment, not just the AI.
3. Small jobs first. One bug, one routine, one folder. Trust is earned in both directions.
4. Treat it like a talented new hire. Clear instructions, low-stakes assignments, check the early work. Autonomy grows with track record — the agent's and yours.

Words you can now use at parties