Your app - Codex and the browser
Conversation on the left, result on the right
In short ↳On this account ChatGPT and Codex are available in the same desktop app. The conversation is on the left, the page preview on the right. An open browser panel does not mean the agent has access to every app or database.
Where to find it: Codex → browser panel beside the chat
Select the screenshot to enlarge it.
What you see on screen
- 1
The '…' menu with tools
Projects, Images, Sites, Maps, GPT, Code review and connected apps (MagicPath, fal, Higgsfield, Figma, Canva, Adobe). A pin means the entry is attached to the left bar.
- 2
Conversation with the agent
You write what to do here, and the agent replies with what it did and what it could not. On the screenshot it honestly says it did not change the database because a plugin was unavailable.
- 3
'Work with Codex' box
Describe the next step. When there is an error, paste the message and say what you did, as shown in the conversation.
- 4
Model and effort (GPT-6 Astra, Light)
The model and effort choice. More on choosing in lesson 13.
- 5
Browser tabs
As in a normal browser: each tab is a separate page. There are three here: the app being built, a project dashboard on the web and a code file. The plus opens another.
- 6
Address bar and buttons
Type an address or paste a link. Arrows and refresh work as usual. The extra buttons beside it help you point at a page element and leave a note for the agent.
- 7
Live preview
What is being built in the project. After each change refresh the tab and compare with the previous version.
- 8
Window layout
The button at the top right shows or hides the browser panel. Hide it when you need more room for the conversation.
How to use it - in plain words
- Open Codex and pick a project (ideally a test folder).
- Describe what should be built and how you will check it is done.
- Open the browser panel beside the chat and enter the preview address the agent gave you.
- Ask for one change, refresh the tab and look at the result. Repeat in small steps.
- When something is off, point at the element on the page and tell the agent what to fix.
- Ask the agent to walk through the page in the browser and check the forms. You see every click and can stop it.
The agent uses this browser within the permissions you grant. Do not sign it in to accounts holding real customer data until you have checked what it can do.
Try it on your account
In a test folder ask Codex for a simple page with one button. Open the browser panel, look at it and ask for a different button colour. See the result without leaving the app.
Your app - tools in tabs
Canva without leaving the app
In short ↳Connected services open in tabs of the same app. You design in Canva while the assistant is next to you in the same window.
Where to find it: Left bar → an app (e.g. Canva)
Select the screenshot to enlarge it.
What you see on screen
- 1
App icon on the bar
A pinned app has its own button. The highlighted icon shows which one you are in.
- 2
'Canva' tab
The app opens like a page in a browser. You can have several tabs at once.
- 3
Plus: new tab
Opens another tab, e.g. a second app alongside.
- 4
Starter suggestions
Ready ideas for a first task, e.g. a resume, lesson plan or video thumbnail. Click one to start from an example.
- 5
Getting started
Example cards, e.g. Instagram posts or turning an image into an editable design. The arrows below scroll to more.
- 6
Button on the right
Opens the assistant panel beside the app so you can ask for help without closing Canva.
How to use it - in plain words
This layout suits tasks where you look at the result: graphics, a presentation, a post. Instead of copying between windows, you work in one.
Start with the starter suggestions, then describe your own project: for whom, in what style, in which format. Check plugin installation separately from the app connection and permissions under Settings → Apps. Opening a Canva tab alone does not confirm agent access to designs.
Try it on your account
Open one of your connected apps in a tab, click a starter suggestion and see what working in a single window feels like.
Module C
Integrations and the desktop app - an agent that reaches for context itself
Three things that turn the desktop app into a work tool: the agent uses the connected tools within the permissions you grant it, it can drive a browser next to the conversation, and it can generate editable motion graphics.
One brief, many tools - with no configuration along the way
It used to take: "open the calendar", then "now the mail", then "now the notes". Today you connect the tools once and describe the outcome. The agent reaches for the sources you point it to and that you have granted it access to - and it is the one stitching the pieces into a single deliverable.
Calendar
who, when, about what
the history of commitments
Meeting notes
what was said last time
Drive
offers, contracts, price lists
Dla zaawansowanych: przykładowy prompt +
Get me ready for tomorrow. DELIVERABLE: a deck - one slide per meeting in tomorrow's calendar. Each slide: company and attendees, what we discussed last time, open commitments with dates, the documents the conversation is about, 3 questions to ask, one risk. SOURCES: decide yourself which connected tools to use. If you cannot find something, write "no data" instead of guessing. CONSTRAINTS: send nothing, change no calendar entries, open no documents outside my company drive. AT THE END: list which sources you actually used.
The last line of that prompt - "list which sources you used" - is the cheapest QA there is. Without it you cannot tell data from your mailbox apart from what the model filled in.
Availability depends on your plan, region, workspace and administrator configuration.
The browser built into the app
Plenty of pages have no API and only render data after a click. The built-in browser lets the agent walk that path like a person: click, scroll, wait for content to load and read what appeared.
Navigation beside you
The browser opens in a panel next to the conversation. You see what the agent clicks and can stop it mid-way - it is not a black box.
Sessions and logins
Tabs can keep a session, so the agent does not log in again at every step. What makes this convenient is also the biggest risk - only sign in where you deliberately want to grant access.
Multiple tabs at once
The agent keeps several pages open and builds one comparison out of them - offers, price lists or specs, with a link behind every row.
Testing your own site
Walking a user path step by step: a form, a cart, a signup. The agent reports what broke, at which step and with what input.
Example: automated form testing
Dla zaawansowanych: przykładowy prompt +
GOAL: check whether the contact form on [URL] works correctly. WALK THE PATH: home page → contact → fill the form with test data ([TEST NAME], [TEST EMAIL]) → submit. ALSO CHECK: empty-field validation, invalid-email validation, the post-submit message, load time of each step, the mobile view. DELIVERABLE: a table of steps - what you did, what was expected, what happened, a screenshot at every difference. End with a bug list ordered by user impact. CONSTRAINTS: our domain only, test data only, do not sign in to production accounts, do not submit the form more than once.
Before you let the agent loose on the web
- Owner's consent - you test what is yours, or what you have written permission to test.
- The site's terms and robots.txt bind the agent exactly as they bind a human.
- Do not bypass logins, paywalls or captchas - if the site guards them, that answers the "am I allowed" question.
- Personal data scraped from pages falls under GDPR just as if you had typed it out by hand.
- Use test accounts, not production ones - agents do sometimes click "delete".
Every action outside the chat
- READthe agent reads
- DRAFTprepares a draft
- REVIEWyou review it
- APPROVEyou approve
- WRITE / SENDonly now it writes or sends
@remotion - motion graphics you can actually edit
The @remotion command generates an animation described in code, not a frozen video file. The side panel gives you a preview player, and you edit the parameters - timings, colours, copy - directly and re-render.
Animation is code
Motion is described programmatically, so you change it with words, not by dragging on a timeline.
Preview beside you
The side player shows the result live - scrub to a frame and see what needs fixing.
Parameters, not rework
Change a scene duration, a colour or a line of text and re-render. You do not restart the project.
A repeatable template
The same animation with swapped data is a template for a whole series - episodes, offers, product cards.
Dla zaawansowanych: przykładowy prompt +
@remotion GOAL: a 20-second opening animation for the "[TITLE]" webinar. SCENES: 1) 0-4 s: the logo slides up, caption "[COMPANY]", 2) 4-12 s: three agenda points appear one after another, 3) 12-20 s: date, time and registration address, ending on a static card. STYLE: colours [HEX 1] and [HEX 2], sans-serif type, calm transitions, no sound effects. 1920x1080, 30 frames per second. IMPORTANT: expose the text and colour parameters as variables so I can change the content without rebuilding the animation.
Rights matter: logos, typefaces, music and photos need licences exactly as in any other tool. The fact that an agent assembled the animation changes nothing legally.
More detailCodex on a phoneDevice filters and projects in the mobile view. The image shows a particular account.
Codex on mobile
Devices and projects: inspect the working context
In short ↳The supplied mobile preview shows device filters and a project list. Selecting a device is not proof that a task is running or that remote access has been granted.
Where to find it: Codex mobile view → device filter → Projects
Select the screenshot to enlarge it.
What you see on screen
- 1
All / device names
All is selected here. MacBook Pro and iMac are filters for the devices shown on this account.
- 2
Green status dots
The visual marks next to devices indicate status. Inspect the application's description; colour alone does not prove a task completed.
- 3
Projects
The visible entries are openai, superassistant and codex. They are examples, not projects automatically created for every user.
- 4
Menu, ellipsis and entry controls
The top buttons and icons next to projects open navigation or actions. Read the menu labels before selecting an operation.
How to use it - in plain words
- Inspect the selected device filter and account.
- Open a project and verify its name and working context.
- Inspect a task result before approving any further action. Do not infer access permissions from this preview.
Try it on your account
Identify the device filters and project list. Explain why a green dot is not a completion report.
