Foundations Module 1 · The AI Landscape

The AI Landscape

Last reviewed

Beginner

What you'll learn

~12 min
  • Tell a stateless chat apart from a workspace agent that works on your project files
  • Recognise the three doors to the same agent: desktop app, terminal CLI, and browser or cloud builder
  • Learn the build loop that drives agentic work: plan, approve, build, check, keep
  • Choose the learning path, and the starting door, that fits your experience and comfort

By the end of this lesson, you’ll know what makes an AI agent different from a chatbot, the three ways to reach one, and which learning path fits you. No coding background and no math beyond everyday arithmetic: we start from zero.

Why this matters

Most professional developers now use AI agents every day (JetBrains, 2026), and the same tools are open to everyone else. This course is not about becoming a programmer. It builds one habit: framing a clear task, directing an agent, and checking what it built. Far fewer people have that habit than have access to a tool.

💬If you're here because you're worried about your job

AI is changing job markets, and if that makes you anxious, you’re paying attention. Plenty of people here are in the same spot: scientists watching grants shrink, analysts seeing dashboards generated for them, developers watching junior roles disappear. What the headlines miss is that your domain expertise is your moat. An agent cannot understand your patients, research questions, or customers the way you do. You’re learning to operate machinery that amplifies what you already know.

🔍Which kind of AI work this isclick to expand

“Working in AI” covers several kinds of work, and much of the anxiety comes from mixing them up:

WhoWhat they doWhat the work asks for
ResearchersInvent new models and ways to train themDeep math and a research setting
ML engineersBuild, evaluate, and run machine-learning systemsStatistics, data pipelines, sizable computing budgets
People who direct finished modelsUse existing models to build and fix real thingsClear thinking and their own domain knowledge

This course is the third row, and it never trains a model. That way of working has a name, orchestrator, and Module 2 defines it.

Stateless chat vs. workspace agents

A plain chat window only knows what you type, paste, or upload, and what it hands back is text you have to put somewhere yourself. Some chat apps can build a single page in the browser (Module 3 shows them), but none of them can open the folders on your computer.

A workspace agent is pointed at a project folder (on your computer or a cloud machine) and works on the files in it directly: it reads, creates, and edits them, runs commands and tests, and fixes its own errors.

Stateless chatWorkspace agent
What it can seeThe conversation and anything you attachThe files in the project folder you opened
What you get backText to copy into files yourselfFiles created and changed in your project
Checking the resultYou run it and report errors backIt runs the code, reads the errors, and tries again
Between sessionsStarts from the chat historyReads the project files and its project notes (CLAUDE.md, AGENTS.md) again

Because these tools change files and run commands, always review what they changed and use version control (you’ll learn Git in Module 7).

If your project folder is a workshop, a stateless chat is a friend on the phone; a workspace agent is that friend standing in your workshop, building alongside you.

Three doors to the same agent engine

Each of the four makers this course covers lets that friend in through three doors. Behind every door is the same agent: it reads the same project files, follows the same project notes, and has the same plan mode, where it asks you questions and proposes a plan before building.

DoorWhat it looks likeWhere it shines
Desktop agent app (a graphical app)Install an app, open a project folder from a menu, type in plain English, review each change on screenStarting without a terminal at all, on Windows, macOS, or Linux
Terminal CLI CLI Command Line Interface — a text-based way to control your computer. Learn more → Open a Terminal Terminal The app/window where you type text commands. Learn more → in your project folder and type one start command: claude, codex, agy, or copilotSpeed, keyboard flow, remote servers over SSH, automation, several agents at once
Browser / cloud builderWork in a web page: build a shareable app in the browser, or hand a GitHub project to an agent on a cloud machineLocked-down work laptops, Chromebooks, quick prototypes

Which door and operating system you start with is your call. Everything this course teaches carries from one door to the next, and the desktop app and the terminal CLI read and write the same folder, so you can switch at any point.

💬Why many developers end up in a terminal

Experienced developers often settle on a terminal CLI on Linux, WSL WSL Windows Subsystem for Linux — runs a Linux environment inside Windows. Learn more → , or macOS: Unix-first build tools run faster, the same commands work over SSH, and tmux lets one person drive several agents side by side. Lesson 4.5 shows that setup, including the course author’s own workflow. Start where you’re comfortable and step up when you want the speed.

The build loop: plan, approve, build, check, keep

Whichever door you use, every piece of work follows the same loop:

  1. Plan: say what you want in plain English, in plan mode. The agent asks questions and proposes a plan.
  2. Approve: read the plan and say yes, or ask for changes.
  3. Build: the agent reads your project, writes the files, and runs them.
  4. Check: open the result and try it. Ask for fixes in plain words.
  5. Keep: save a checkpoint of the version that works (Module 7 shows how).

Your job is to think clearly, give good instructions, and check results. Coding skill multiplies that, but it’s not required to start.

What a session looks like

Here is a short session in Claude Code’s terminal CLI. The desktop apps show the same conversation in a window.

SESSION Claude Code in a terminal: a contact form, start to finish
  1. Typed in the terminal: claude --dangerously-skip-permissions
  2. Mode:Plan mode (Shift+Tab)
  3. You type in the session: I need a contact form for a small business site: name, email, and a message.
  4. The tool did:Read the project folder
  5. The tool replies: A couple of questions first: 1. Should the form send an email, or show a thank-you message for now? 2. Any colours or a logo to match?
  6. You type in the session: A thank-you message for now. Dark theme, clean and simple.
  7. The tool replies: Plan: - contact.html with name, email, and message fields - styles.css with a dark theme - a check that the email looks valid, then a thank-you message Shall I go ahead?
  8. You type in the session: Yes, go ahead.
  9. The tool did:Created contact.html
  10. The tool did:Created styles.css
  11. The tool replies: Done. Open contact.html in your browser: an email without an @ shows a warning, and a valid one shows the thank-you message.
  12. You type in the session: Make the thank-you message green and clear the fields after sending.
  13. The tool did:Updated contact.html
  14. The tool replies: Updated. Refresh the page and send the form again to check.

Illustrative session; the agent's wording varies from run to run. The start line turns the approval prompts off, and the next lesson explains what that setting does.

Notice what didn’t happen: the person didn’t learn HTML, install frameworks, or read documentation. They described what they wanted in a sentence, answered two questions, and refined the result with a follow-up. That’s orchestration, and it looks the same in all four tools and through all three doors.

⚠What if the AI gives a wrong answer?

AI tools sometimes produce wrong code, made-up facts, or results that look right but hide bugs. Always test what the agent builds: run it, click through it, check the output. You don’t need to read every line, but you do need to confirm it works (Module 8 goes deeper). Treat bold AI claims, including ours, the same way.

The four ecosystems you’ll learn

As of November 2026, these are the leading agentic coding tools. Each comes as a terminal CLI and as a desktop or cloud app built on the same agent. Features and pricing change often, so check the latest documentation.

EcosystemMakerTerminal CLI and its appWhat makes it special
Claude CodeAnthropicClaude Code CLI & Claude Desktop (Code tab)Deep codebase understanding, long context
AntigravityGoogleAntigravity CLI & Antigravity AppFree tier in the app; layouts for knowledge work and for software
CodexOpenAICodex CLI & ChatGPT Desktop (Codex)Agent-first design; the desktop app has a free model
CopilotGitHub / MicrosoftCopilot CLI & GitHub Copilot AppGitHub integration; Copilot Free includes the CLI and the app

You don’t need to pick just one. Good orchestrators know which tool, and which door, to reach for.

🧬In Your Field: Biotechclick to expand

Lab scenario: describe the heatmap you need from your RNA-seq CSV to an agent in your project folder, then refine labels and colours. The Biotech track works this way.

📊In Your Field: MIS / Businessclick to expand

Business scenario: describe a quarterly sales dashboard and get an interactive prototype with charts and filters. The Business track shows you how.

🏛️In Your Field: Government Devclick to expand

Government scenario: describe an internal case intake tool and get a prototype for internal review; production still needs security review and ATO.

Where you stand right now

Among adults aged 15 to 64 worldwide, 18.8% used a generative AI product in June 2026 (Microsoft’s telemetry-based estimate), and nine in ten professional developers use an AI coding agent every week (JetBrains, 2026). These measure different groups, so read them side by side, not as one funnel. None of them measures how well the work is set up: a defined task, a place to run, and a way to check the result. That practice is what this course teaches.

AI ADOPTION IN 2026

Three different groups, each counted by a different source. The numbers do not add up or nest.

The world

8.3B people alive in 2026 (UN projection)

  • 2.0B not using the internet, about 24% of the world
  • 18.8% of adults aged 15 to 64 used a generative AI product by mid-2026

The 18.8% is Microsoft’s estimate from usage data, measured against working-age adults. It is a separate measure, not a share of the people online.

Everyday workplaces

30% of U.S. employees use AI at work a few times a week or more

  • 25% say their organization has communicated a clear plan for using AI

Two separate questions put to all employees: the 25% is not a share of the 30%.

Software teams

90% of professional developers surveyed use AI coding agents at work every week

  • 68% use them every day (they are counted in the 90% too)

The verification gap

In METR’s 2025 trial, experienced developers using early-2025 AI tools took 19% longer on tasks in their own projects, while believing the tools had sped them up. METR’s February 2026 follow-up (57 developers, 800+ tasks) points to developers being faster with today’s tools, but METR calls its own data “an unreliable signal” of the current effect.

Having the tool is now common. Saying clearly what you want, and checking what comes back, is what separates real gains from rework.

🔧

Common first-timer concerns

Use the Symptom → Evidence → Request pattern: describe what you see, paste the error, then ask for a fix.

Symptom
I don't know what a 'terminal' or 'CLI' is
Evidence
You see references to 'command line' and 'terminal' and feel lost
What to ask the AI
"That's normal, and you don't need one to start: a desktop agent app works without it. When you're ready, Module 4 teaches the terminal from zero, on Windows, macOS, or Linux."
Symptom
I'm worried I'll break something on my computer
Evidence
You've heard that AI agents and terminal commands can be dangerous
What to ask the AI
"You work inside a dedicated practice folder; Lesson 1.2 explains how approval settings work, including which settings let an agent act without asking. Everything you build in this course lives in its own folder you can delete anytime."

Choose your path

1. Never opened a terminal? You’re in the right spot. In the next lesson you install a desktop app or a terminal CLI and run your first build, then work through Modules 1 to 11 in order. The desktop app needs no terminal at all.

2. Comfortable in a shell but new to AI agents? Consider the developer track. It moves faster through the basics and spends more time on agentic patterns.

3. Already running Claude Code or Codex daily? Modules 1 to 11 will read as review. Go to Module 20: The Prompting Playbook and Module 19: Agentic Engineering for advanced prompting, MCP servers, skills and hooks, multi-agent work, and headless pipelines. Module 7 (Git), Module 8 (Understanding What AI Builds), and Module 11 (Ship It) are still worth a skim.

Key takeaways

  • Stateless chat vs. workspace agent: a chat sees only the conversation; a workspace agent reads, creates, edits, and runs the files in your project folder
  • Three doors, one engine: desktop agent app, terminal CLI, or browser / cloud builder; the skill carries across all of them
  • The build loop: plan, approve, build, check, keep. Plain English in, working software out
  • Four ecosystems (Claude Code, Antigravity, Codex, Copilot), each with a CLI and an app
KNOWLEDGE CHECK

What makes a workspace agent different from a stateless chat?

Next step: In the next lesson you install a desktop agent app or a terminal CLI and run your first build on your own computer.

Search lessons