Learn every AI tool, one card at a time

ChatGPT lessons

ChatGPT is made by OpenAI (United States). The deck has 79 cards in five levels, from Rookie to Legendary. Every card was checked against current sources on 2026-11-05.

Rookie

1. Give it a clear assignment (Rookie, 6 min). Say what you want, who it is for, what it should know and what counts as done. That beats any clever prompt trick.
2. Show it what good looks like (Rookie, 5 min). Paste one example you like, or describe the finished answer: length, layout, tone, what it must include. An example often shows more than a paragraph of instructions.
3. Get answers you can paste straight in (Rookie, 4 min). Name where the answer is going: a Slack message, a slide, a spreadsheet column. Give the length and format, and you skip the clean-up.
4. Rewrite anything for a different reader (Rookie, 4 min). The same facts land differently for a CEO, a customer and a new hire. Tell ChatGPT who will read it and what they need to do next.
5. Chat, Work or Codex: pick the right place (Rookie, 5 min). Chat is for questions, Work is for finished files and long tasks, Codex is for code. Starting in the right place gets you the right model and tools.
6. Choose how hard it thinks (Rookie, 5 min). Fast mode for simple questions, more thinking for hard ones. Matching the two saves you time and gets better answers.
7. Improve the answer instead of starting over (Rookie, 5 min). The first answer is a draft. Short, specific follow-ups get you to a great result faster than a perfect first prompt.
8. Check what it tells you (Rookie, 6 min). Ask for sources, open them, and check every number you plan to repeat. Thirty seconds of checking protects your name.
9. Ask it to do things, not whether it can (Rookie, 4 min). ChatGPT cannot see its own systems. So skip “do you have access to memory?” and just say “save this to memory”. Trying is the only real test.
10. Make it search the web and show sources (Rookie, 5 min). For anything that changes, such as prices, news, rules or opening hours, turn on Search and click at least one source before you trust the answer.
11. Use a temporary chat for one-off questions (Rookie, 4 min). A temporary chat stays out of your history, is not used to train models and never updates your memory. Pick it for the question you do not want following you around.
12. Tune its style once, not in every chat (Rookie, 4 min). Characteristics let you dial warmth, enthusiasm, formatting and emoji up or down for every chat. Set them once and stop typing “be shorter” forever.
13. Ask for the answer as a chart, table or tool (Rookie, 5 min). GPT-6 can answer with charts, side-by-side comparisons, buttons and small interactive tools. Say which shape you want and it builds it in the chat.
14. Learn anything with study mode (Rookie, 5 min). Study mode teaches instead of answering. It asks what you know, explains in layers and quizzes you, so the knowledge ends up in your head and not just the chat.
15. Turn your notes into flashcards (Rookie, 4 min). Ask for flashcards on a topic or upload notes. You get a deck you can flip, mark as known or not, shuffle and reopen later from your library.
16. Think before you click its links (Rookie, 3 min). A link from ChatGPT can carry details from your chat to another website. Read the address before you click, and be extra careful with links you are asked to approve.
17. Spend your usage where it counts (Rookie, 5 min). Bigger models and longer tasks use more of your allowance. From December, Instant chat will use none of it. Save the heavy settings for the work that needs them.
18. Keep your data under control (Rookie, 5 min). Know three switches: model training, temporary chat and locked chats. Set them once and you can use ChatGPT without second-guessing.
19. Find any old chat, file or project fast (Rookie, 3 min). Sidebar search finds chats, projects, images and files by words you remember. For chats it matches titles and messages. On the web, Cmd+K or Ctrl+K opens it instantly.
20. Boss: your first real deliverable (Rookie, 10 min). Prove the Rookie skills on one real piece of work: a clear assignment, an example, a paste-ready format, one round of improvement and a fact check.

Apprentice

21. Set it up once: instructions and memory (Apprentice, 8 min). Put how you want answers in custom instructions, facts about you in memory, and task material in projects. Then stop repeating yourself.
22. Write custom instructions that change answers (Apprentice, 6 min). Custom instructions apply to all your chats. Use them for facts about your work and rules you would otherwise repeat daily, not vague wishes.
23. Take control of what it remembers (Apprentice, 6 min). Depending on plan and region, Memory can use past chats, saved memories, custom instructions, your files and connected apps. Review the summary, correct what is wrong, and tell it what never to bring up again.
24. Work with your own files and recordings (Apprentice, 7 min). Upload the spreadsheet, PDF or meeting recording and ask your question about it. No copying and pasting.
25. Run ongoing work in a project (Apprentice, 8 min). Give every ongoing piece of work its own project with files and instructions. Each new chat inside it already knows the background.
26. Edit drafts right inside the chat (Apprentice, 5 min). Emails, posts and documents land in writing blocks you can edit directly. Select one sentence, ask for a change and keep the rest exactly as it is.
27. Talk to it and make images (Apprentice, 7 min). Talk through ideas out loud with Voice, and create or edit images from a sentence, a template or even a sketch (sketches on mobile).
28. Make and fix images without starting over (Apprentice, 6 min). Describe the image in detail, then fix it in place: select an area and say what to change. Templates, any aspect ratio and transparent backgrounds are built in.
29. Make a looping GIF and reuse a saved photo (Apprentice, 6 min). Ask for a loop and say exactly what moves. Save a photo you use often as a reference, then call it with @ and its name in any later chat.
30. Talk through an idea, get back a plan (Apprentice, 5 min). Hit Record, think out loud for five minutes, and ChatGPT turns it into a transcript and summary you can make into a plan, an email or a brief.
31. Analyse a spreadsheet the right way (Apprentice, 6 min). Clean the sheet first (one header row, one record per row), say what you want to learn, and read the code it runs before you trust the number.
32. Use ChatGPT inside Excel and Google Sheets (Apprentice, 5 min). ChatGPT now sits in a sidebar inside Excel and Google Sheets. It can build a sheet from scratch, explain a big workbook and change formulas while you watch.
33. Draft and fix documents inside Word (Apprentice, 5 min). ChatGPT for Word works on the document you have open. Select text, say what to change and what to keep, and ask for a plan before big edits.
34. Use deep research for a real question (Apprentice, 10 min). For questions that need many sources, let ChatGPT plan the research, read for you and bring back a cited report.
35. Make it check its own work (Apprentice, 7 min). Before you accept an answer, have ChatGPT grade it against your checklist and fix what fails. Quality goes up without more of your time.
36. Lock your sensitive chats (Apprentice, 4 min). Locked chats hide chosen conversations behind a 4-digit PIN, so they stay out of your history and search when you share your screen.
37. Boss: make it know your job in 30 minutes (Apprentice, 30 min). Set up ChatGPT so a fresh chat already knows your role, audience, formats and key documents. Beat it by getting a great answer from a one-line prompt.

Pro

38. Connect your apps with plugins (Pro, 8 min). Connect the apps your work lives in, then ask questions across them. Your inbox, calendar and files become part of the conversation.
39. Decide how much your apps may do alone (Pro, 5 min). For each connected app, choose: always ask, allow reading, allow low-risk actions, or allow everything. Start strict and loosen only what you trust.
40. Turn a repeated task into a skill (Pro, 9 min). On Business, Enterprise and Edu workspaces: if you explain the same workflow every week, build it once as a skill. ChatGPT then uses it automatically when it fits.
41. Move your custom GPTs before 11 December (Pro, 8 min). Custom GPTs retire on 11 December 2026. If you built one, migrate it to a plugin as soon as the option appears in your account and test it before the date.
42. Hand a whole deliverable to ChatGPT Work (Pro, 10 min). For finished documents, spreadsheets and decks, use Work. Give it the sources, the format and what must not change.
43. Pick the right model for each job (Pro, 5 min). Use GPT-6.1 Terra for everyday writing, forms and computer tasks, GPT-6.1 Sol for complex coding and research, and GPT-6 Astra only for the hardest problems.
44. Write a brief an agent cannot misread (Pro, 6 min). Before handing over a task, write four things: the result, what it may decide alone, when it must stop and ask, and what “done” includes.
45. Let ChatGPT work on websites with you (Pro, 6 min). Open the built-in browser, sign in yourself, and let ChatGPT do the clicking. On sites that offer site tools, it can use them directly and work faster.
46. Put recurring work on a schedule (Pro, 7 min). Anything you ask for every Monday can run by itself. Schedule it, or let it start when an email or Slack message arrives.
47. Borrow from the best prompt libraries (Pro, 7 min). Study what thousands of power users share on GitHub, then adapt it to your context. Copying a prompt word for word rarely gets their results.
48. Keep your work in Space, not scattered chats (Pro, 6 min). Space holds pages: living documents you and ChatGPT keep editing. Turn the useful parts of chats into pages and stop digging through history.
49. Build a deck with ChatGPT and your team (Pro, 6 min). Ask Work to “create ChatGPT slides” from your notes and files, guide it with a template, then edit the same deck together in real time.
50. Let the Meetings plugin take your notes (Pro, 5 min). The Meetings plugin takes notes on calls and in-person meetings without a bot joining, then saves a summary with action items in Space.
51. Let Codex rebuild your Excel workbook (Pro, 6 min). Codex can drive an open Excel workbook through the add-in: sheets, formulas, formatting and charts, plus your local files and code, in one task.
52. Work with @ChatGPT in Slack and Teams (Pro, 5 min). Mention @ChatGPT in a channel or thread to summarise, draft or answer with your team, without anyone leaving the conversation.
53. Use your ChatGPT plan inside other apps (Pro, 4 min). In participating apps, sign in with ChatGPT and let them use your plan for AI requests. Set a weekly limit per app so nothing eats your allowance.
54. Boss: hand over a whole weekly report (Pro, 30 min). Turn one report you write every week into a hands-off routine: connected sources, a skill for the format, a schedule, and a review that takes two minutes.

Master

55. Delegate like a manager: boundaries and done (Master, 9 min). For long tasks, say what it may decide alone, when it must ask you, and what finished looks like. Then step back.
56. Give agents the evidence, not the noise (Master, 6 min). More context is not always better. Cut what the task does not need, keep the evidence it does, and put stable material first.
57. Steer a long task without starting over (Master, 5 min). When plans change mid-task, steer: say what should change and what must stay. Answer its questions fast, and say which work can continue while you decide.
58. Make agents prove their work (Master, 6 min). Define done as “made it, ran it, checked it, fixed it”, and require a handoff that says what was checked and what still needs you.
59. Split big jobs across parallel agents (Master, 8 min). For big, many-sided tasks, describe the teams you want. GPT-6 Pro can run several agents at once and combine what they find.
60. Build your weekly AI system (Master, 12 min). Combine a project, a template or skill, and a schedule. Your recurring work then starts itself, follows your format and lands ready to review.
61. Meet your dot: an agent that keeps working (Master, 7 min). A dot is an always-on agent with its own cloud computer, your memory and your connected apps. Give it a project, not a question, and it keeps going between chats.
62. Set your dot’s rules before it acts (Master, 6 min). Your dot shares plugin permissions with ChatGPT and Codex. Add Custom Rules such as “never send emails”, and know what its proactive research can and cannot do.
63. Hand recurring team work to a team task (Master, 6 min). A team task runs in the cloud on a schedule or trigger under the team’s service account, and any authorised teammate can fix it. Recurring work stops depending on one person.
64. Start work automatically when something happens (Master, 5 min). Plugins can now start automations when something happens in a connected app, like a new task on a board. ChatGPT reads the context and drafts the next step while you are away.
65. Ask your company data real questions (Master, 6 min). The Data plugin connects to your business data, investigates what changed, explains it and builds interactive dashboards you can refine in the same chat.
66. Share pages without leaking your memory (Master, 5 min). Sharing a page never shares your chats or Memory. But anything ChatGPT wrote onto the page from your memory is now visible to everyone who can read it.
67. Stay safe when ChatGPT acts for you (Master, 6 min). When ChatGPT can read your email and click buttons, a hidden instruction on a web page can try to hijack it. Limit access and keep the high-impact steps for yourself.
68. Turn on Lockdown Mode for sensitive work (Master, 5 min). Lockdown Mode limits tools that reach the web and outside services, which makes it much harder for a prompt injection to send your data out. It costs features, so use it where data is sensitive.
69. Boss: three agents, one week, zero chaos (Master, 30 min). Run three independent long tasks in parallel for a week, each with a four-part brief, lean context, steering when needed and a proven handoff.

Legendary

70. Work the way OpenAI itself works (Legendary, 7 min). Inside OpenAI, every department, including Legal and Recruiting, now uses Codex as its main AI tool. The heaviest users run several agents in parallel instead of chatting.
71. Draw your AI org chart (Legendary, 8 min). Map every recurring job to the right layer: chat for one-offs, pages for shared context, skills for how-to, tasks for routines, a dot for ongoing goals, Codex for long builds.
72. Mine the giant prompt libraries, then rewrite (Legendary, 6 min). The biggest prompt collections on GitHub are a goldmine of ideas, but many were written for older models. Take the idea, rewrite it as a clear GPT-6 assignment, and save the winners as skills.
73. Package your workflow as a plugin (Legendary, 8 min). A plugin bundles skills, app connections and tools into one installable package. Start with Plugin Creator and copy the structure of OpenAI’s own examples on GitHub.
74. Run your team’s plugins from GitHub (Legendary, 6 min). Admins can import a plugin marketplace (a JSON catalog in a GitHub repo) into the workspace. Daily sync keeps every plugin up to date as the repo changes.
75. Plug your own system into ChatGPT (Legendary, 7 min). Any internal tool can become a ChatGPT tool through an MCP server. Admins test it privately, then publish it, and the Apps SDK can add an interface in the chat.
76. Build an agent that runs your process (Legendary, 7 min). The Agents API gives developers Codex’s machinery (multi-agent work, tool search, context compaction and now computer use) while OpenAI runs the infrastructure.
77. Teach it: turn your team into ChatGPT pros (Legendary, 7 min). Legends multiply. Run a one-hour workshop on your team’s real tasks, share your skills and plugins, and point people to OpenAI Academy’s free courses for the basics.
78. Brief every coding agent with one AGENTS.md (Legendary, 12 min). Write your project’s rules once in AGENTS.md, add folder-level files where teams work differently, and check which files Codex actually loaded.
79. Boss: the 30-day Legendary quest (Legendary, 8 min). Four weeks, four layers: a skill, a routine, a connected system and a delegated goal. By day 30, ChatGPT runs part of your work while you review it.

Copilot lessons

Copilot is made by Microsoft (United States). The deck has 38 cards in five levels, from Rookie to Legendary. Every card was checked against current sources on 2026-11-05.

Rookie

1. Know which Copilot you are signed in to (Rookie, 4 min). One Copilot app serves personal and work accounts. The label next to your profile tells you which one you are in, and the two never share data.
2. Pick the right Copilot plan (Rookie, 5 min). Free covers chat, images and uploads within limits. Microsoft 365 plans add Copilot in the Office apps, and only Premium and Pro add Cowork and agentic research.
3. Write a prompt Copilot can act on (Rookie, 5 min). Say what you need, give the details Copilot cannot guess, point it at the source and ask for the format you want.
4. Point Copilot at the exact file (Rookie, 5 min). Attach or reference the one file that matters. For long documents, ask about one section at a time, because the middle of a long file gets less attention.
5. Check where an answer came from (Rookie, 4 min). Open Sources to see the exact Bing query Copilot used and the pages it read, then confirm anything important yourself.
6. Catch up on a long email thread in Outlook (Rookie, 4 min). Select Summary by Copilot at the top of a thread, then use the numbered citations to jump to the email behind each point.
7. Control what Copilot remembers about you (Rookie, 5 min). Saved memories, chat history and (on personal accounts) shared experiences and ad personalisation each have their own switch. Turning memory off does not delete what is already saved.
8. Boss: your first real deliverable with Copilot (Rookie, 15 min). Take one real task from this week, use the right account, give Copilot the file and a clear brief, check the sources, and send a finished draft.

Apprentice

9. Set custom instructions once (Apprentice, 5 min). Tell Copilot once how you work and how you want answers formatted, and it applies that to every future chat.
10. Get decisions and owners out of a Teams meeting (Apprentice, 6 min). Set Allow Copilot and Facilitator before the meeting, keep transcription on if you want to ask questions afterwards, and ask for decisions, owners and dates.
11. Draft and edit inside Word (Apprentice, 6 min). Open Copilot from the button in the corner of your document, describe the outcome, and use Chat only when you want answers without changes to the file.
12. Use edit, plan and chat modes in Excel (Apprentice, 6 min). Plan mode shows the steps before anything changes, edit mode builds in the workbook, and chat mode answers questions without touching your data.
13. Turn a good answer into a Copilot Page (Apprentice, 5 min). Move a response into Pages to keep editing it with Copilot and your colleagues, instead of copying text into a document.
14. Run deep research with Researcher (Apprentice, 8 min). Researcher takes minutes instead of seconds and returns a structured report with citations. Answer its clarifying questions and tell it whether to use work data, the web or both.
15. Build a deck from a Word file in your company’s look (Apprentice, 7 min). Start from your company template or attach a deck whose look you want, pick the Brand Kit if your organisation has one, then describe the deck.
16. Boss: a meeting-to-document workflow (Apprentice, 15 min). Take one real meeting or project, pull the decisions out with Copilot, research the open question and deliver a finished Word document or Page.

Pro

17. Hand a multi-step task to Cowork (Pro, 8 min). Chat answers. Cowork does the work: it plans the steps, creates and edits files, drafts and sends email, and stops to ask before sensitive actions.
18. Approve Cowork actions the safe way (Pro, 5 min). Read every approval card before you select the action button. Use the option to skip future prompts only for actions you would be happy to see repeated without you.
19. Put routine work on a schedule or a trigger (Pro, 7 min). Say the timing in plain words for a schedule, or describe the event for a trigger. Triggers default to draft-and-approve, but a schedule may ask for its approvals at setup so it can run unattended, so approve only what you would let run without you.
20. Decide when Copilot should use your work data (Pro, 5 min). Work IQ on lets Copilot use your mail, files, chats and meetings. Work IQ off gives a general answer from the web, your attachments and your profile.
21. Keep ongoing work in Projects and Workspaces (Pro, 6 min). Projects are for your own ongoing work. Workspaces are shared spaces for a team. Both keep files, instructions and context together so you stop rebuilding it every chat.
22. Analyse data files with the Analyst agent (Pro, 7 min). Attach the files, ask a clear question about your data, and get back a report with statistics, trends, outliers and charts.
23. Let Cowork finish the web steps in your own Edge (Pro, 8 min). Cowork can do the website parts of a task in a hidden tab of your own Edge, with your sign-ins, and hands the tab back to you for passwords, multifactor codes and CAPTCHAs.
24. Boss: delegate a real weekly task end to end (Pro, 20 min). Pick one recurring task, run it once with Cowork and approvals, then schedule it in draft mode and keep it in a Project.

Master

25. Keep agents safe from hidden instructions (Master, 7 min). Treat web pages, emails and shared files as content, never as commands. Watch for unexpected sites or approvals, and pause the task the moment something looks off.
26. Write your own Cowork skill (Master, 10 min). A skill is a SKILL.md file of instructions that Cowork loads when the task fits. Write one for any process you repeat, and test it in a new session.
27. Build a team agent with Agent Builder (Master, 10 min). Describe the agent in plain language, give it approved knowledge, test it, then share it. Check first whether an existing approved agent already does the job.
28. Know what enterprise data protection covers (Master, 8 min). Prompts and responses get the same contract terms as your Exchange mail and SharePoint files and are not used to train foundation models. Web queries to Bing, oversharing and some model choices sit outside that promise.
29. Understand licences, credits and usage-based billing (Master, 8 min). A user licence covers everyday Copilot. Long-running agentic work such as Cowork, Code and Autopilot runs on usage-based billing with Copilot Credits, so set spending limits before you scale.
30. Boss: ship a safe team setup (Master, 25 min). Build one shared skill or agent for your team, with trusted knowledge, an owner, data checks and a spending limit, and prove it on a real task.

Legendary

31. Choose how to build: Agent Builder, Copilot Studio, Agents Toolkit or the Agents SDK (Legendary, 12 min). Declarative agents use Copilot’s own models and orchestrator. Custom engine agents bring your own. Pick the lightest tool that meets the need.
32. Give a Copilot Studio agent tools through MCP (Legendary, 12 min). Connect an MCP server and every tool and resource it publishes appears in your agent and stays in sync. Turn on generative orchestration first.
33. Bring outside data into Copilot with connectors (Legendary, 10 min). Synced connectors index outside data into Microsoft 365 so Copilot and search can find it. Federated connectors fetch it live through MCP without copying it.
34. Use Copilot inside your own apps with the Copilot APIs (Legendary, 12 min). The Retrieval API grounds your own model in Microsoft 365 content without copying data out. The Chat API (preview) brings Copilot conversations into your apps.
35. Build a custom engine agent with the Agents SDK (Legendary, 14 min). The Agents SDK handles channels, messages and conversation state. You bring the model and the logic, in C#, JavaScript or Python.
36. Learn from Microsoft’s Copilot GitHub repos (Legendary, 10 min). Microsoft’s official repos show working MCP servers, agent samples, hands-on labs and prompt libraries. Read them before you build from scratch.
37. Build a declarative agent as code with Agents Toolkit (Legendary, 14 min). When an agent matters to a team, build it in VS Code: its instructions and conversation starters live in files you can review in Git and provision again in one click.
38. Boss: design a Copilot system for your team (Legendary, 30 min). Design and ship one system that connects Copilot to an outside source, acts through a tool, is governed by an owner and a budget, and proves its value in numbers.

Gemini lessons

Gemini is made by Google (United States). The deck has 48 cards in five levels, from Rookie to Legendary. Every card was checked against current sources on 2026-11-05.

Rookie

1. Give Gemini a clear brief (Rookie, 5 min). Say the goal, give the background, show one example and name the format. Gemini keeps answers short unless you ask for more.
2. Pick the model and thinking level for the job (Rookie, 4 min). Flash-Lite for quick asks, Flash for everyday work, Pro or Argon for hard problems. Raise the thinking level only when the task needs it.
3. Give Gemini the whole file, then ask (Rookie, 5 min). Upload up to 10 files in one prompt, put your question after them, and ask for answers that point back to the page or cell.
4. Check where an answer came from (Rookie, 4 min). Open Sources under an answer, check the claims you will use, and correct or regenerate when something is off.
5. Make and edit images with Nano Banana (Rookie, 5 min). Describe the subject, the setting, the style and any exact words. Then edit by chatting, one change at a time.
6. Check if an image, video or audio was made with Google AI (Rookie, 4 min). Upload the file and ask “Was this created or edited by Google AI?”. Gemini checks for SynthID watermarks and Content Credentials.
7. Draft in Canvas, not in the chat (Rookie, 5 min). Open Canvas for anything you will edit: docs, slides, code or a small app. Select a part and ask for a change instead of regenerating everything.
8. Talk it through with Gemini Live (Rookie, 4 min). Speak naturally, interrupt when you like, and turn on your camera or share your screen so Gemini sees what you mean.
9. Decide what Google keeps from your chats (Rookie, 5 min). Keep Activity is on by default for adults. Turn it off, or use a temporary chat, for anything you would not want a reviewer to read.
10. Share and export answers the right way (Rookie, 4 min). A shared chat is a public link anyone can open. For colleagues, export to Docs, Gmail or Sheets instead.
11. Make your usage last (Rookie, 4 min). Limits are based on compute and refresh every 5 hours up to a weekly cap. Media, Deep Research, Pro and extended thinking use the most.
12. Boss: your first real deliverable with Gemini (Rookie, 15 min). Take one real task from this week, give Gemini the files and a clear brief, check the sources, and export a finished draft.

Apprentice

13. Tell Gemini how to answer you, once (Apprentice, 5 min). Add standing instructions in Settings, then Personal context, so every chat starts with your preferences.
14. Turn a repeated prompt into a skill (Apprentice, 6 min). Save a task you repeat as a skill with a clear name and a “Use when” description. Gemini then applies it by itself when a prompt matches.
15. Move your Gems to skills before 17 December (Apprentice, 6 min). On 17 November 2026 Gems become read-only, and on 17 December they disappear. Google converts them for you, but some files do not make the move.
16. Make memory work for you (Apprentice, 5 min). Turn Memory on so Gemini learns from past chats. Ask what it used, correct it in the chat, and delete chats to make it forget.
17. Bring your memory and chats from another AI (Apprentice, 5 min). Paste Gemini’s import prompt into your old AI, paste its reply back, and upload your exported chat history as a .zip.
18. Connect Gmail, Drive and Calendar (Apprentice, 5 min). Connect the Google Workspace app, then type @ and pick Gmail, Drive or Calendar so Gemini searches the right place.
19. Use Deep Research for questions that need many sources (Apprentice, 6 min). Write the question, fix the research plan before it starts, add your own sources if needed, and turn the report into a doc or an Audio Overview.
20. Keep a project in a notebook (Apprentice, 5 min). Create a notebook for each ongoing project. It remembers its sources, instructions and past chats, and syncs with Gemini Notebook.
21. Boss: make Gemini know your job in 30 minutes (Apprentice, 15 min). Set your instructions, build two skills for your most repeated tasks, connect Workspace, and run one Deep Research report for a real decision.

Pro

22. Hand Gemini a task, not a question (Pro, 6 min). Switch to Task, describe the outcome, the deadline and the limits, and let Gemini work across your apps, files and the web while you do something else.
23. Put a task on a timer or a trigger (Pro, 6 min). Schedules run a task at set times, when a matching email arrives, or when a topic you follow changes. Test each one with Run now.
24. Schedule a daily briefing in chat (Pro, 4 min). Ask for a recurring digest in plain words. Gemini prepares it in the hour before your delivery time and notifies you.
25. Decide what Gemini may do without asking (Pro, 5 min). Use Allow once for new actions, Always allow only for safe and reversible ones, and keep sending and deleting on Always ask.
26. Connect your Google apps with Personal Intelligence (Pro, 5 min). Choose which apps Gemini may learn from: Workspace, Photos, YouTube, Search, Contacts or Wallet. Turn it off for any chat that should stay generic.
27. Connect the other apps you use (Pro, 4 min). Open Customize, then Connectors, add the apps you trust, and type @ plus the app name to make Gemini use it.
28. Use Gemini on your desktop (Pro, 4 min). Press Option+Space on a Mac or Alt+Space on Windows to open Gemini over any app. On a Mac, hold fn to talk to the window you are working in.
29. Make a short video with Gemini Omni (Pro, 5 min). Describe the shot, add up to 5 images and one video as references, then edit by chatting: remove objects, change the camera angle or change the scene.
30. Boss: automate one weekly chore end to end (Pro, 15 min). Pick a weekly chore, build it as a task with a skill and a schedule, set permissions carefully, and run it for two weeks with checks.

Master

31. Set up agents so mistakes stay small (Master, 6 min). Give agents the least access they need, keep sensitive actions on Always ask, keep secrets out of threads, and treat outside content as untrusted.
32. Let Gemini work on files on your Mac (Master, 6 min). Add only the folders a task needs under Connected folders, describe the outcome, and check the result before you start your next task.
33. Connect your own tools with an MCP server (Master, 6 min). Paste a trusted MCP server URL under Custom apps, then call it with @. Google does not secure third-party servers, so check what it can do first.
34. Build a skill package with files (Master, 6 min). Put a SKILL.md and its reference files in one folder, keep it under 100 MB with supported file types, and upload it. Skills built for other platforms upload the same way.
35. Know when Argon is worth it (Master, 5 min). Argon is Google’s top model for long, complex work. Use it with a high thinking level for hard analysis and big builds, and Flash for everything else.
36. Build an app with Antigravity in the Gemini Mac app (Master, 6 min). Click Antigravity, describe the app, review and comment on the plan before the agent changes anything, and read its walkthrough at the end.
37. Boss: three automations, one week, zero surprises (Master, 20 min). Run three agent workflows for a week (a Mac files task, a Gmail-triggered task and one connected or custom app) with least access, Always ask on risky actions, and a written log.

Legendary

38. Your first Gemini API call with the Interactions API (Legendary, 10 min). Get a key in Google AI Studio, install google-genai, and call client.interactions.create. Google recommends the Interactions API for all new projects.
39. Cut API costs with caching, Batch and the right model (Legendary, 10 min). Put long shared content first for implicit cache hits, send non-urgent work through the Batch API at half price, and plan for the price rises on 1 December and 1 January.
40. Ground answers in Google Search and your own files (Legendary, 10 min). Add the google_search tool for fresh facts with citations, and use File Search to answer from your own documents.
41. Make every answer come back in the same shape (Legendary, 8 min). Give Gemini a JSON Schema (or a Pydantic or Zod model) and get predictable, typed output for extraction, classification and agent tools.
42. Choose the right Google tool for coding agents (Legendary, 8 min). Since 18 June 2026, personal and Google AI plan users get their terminal coding agent from Antigravity CLI (agy). Gemini CLI now serves enterprise Code Assist, and gemini-api is a plain API client for scripts and other agents.
43. Move your app off Gemini 2.5 before 1 December (Legendary, 8 min). From 1 December 2026, calls to gemini-2.5-pro go to gemini-3.8-flash and calls to gemini-2.5-flash go to gemini-3.5-flash-lite, at the new models’ prices. Test now.
44. Learn from Google’s own Gemini code on GitHub (Legendary, 8 min). Start with the Gemini Cookbook for working examples, the google-genai SDKs for the API and the Agent Development Kit for agent systems. Check the model and package names before you copy.
45. Connect Gemini to your own tools with function calling (Legendary, 12 min). Describe your functions. Gemini returns the function name and arguments, and your code runs it and sends the result back.
46. Build a voice agent with the Live API (Legendary, 14 min). The Live API streams voice in and out over one WebSocket, so people can talk to your agent and interrupt it like a person.
47. Give your app its own voice with Gemini TTS (Legendary, 12 min). Design a voice from a short description, remix it with small relative edits, and keep each line’s mood in style so the speaker stays the same.
48. Boss: the 30-day Legendary quest (Legendary, 15 min). In 30 days, ship one system from each quest line: a skill package, a connected tool, an API app with structured and grounded answers, and a team rollout.

Claude lessons

Claude is made by Anthropic (United States). The deck has 65 cards in five levels, from Rookie to Legendary. Every card was checked against current sources on 2026-11-05.

Rookie

1. Brief Claude like a brilliant new colleague (Rookie, 5 min). Claude knows a lot but knows nothing about you. Say who it is for, what you need, the format and any limits.
2. Give Claude the whole document, then ask (Rookie, 6 min). Do not summarise it first. Upload the full document, put your question after it, and ask for quotes you can check.
3. Say why, not just what (Rookie, 4 min). Give the reason behind each rule. Claude follows a rule better when it knows what the rule protects.
4. Show one example of good (Rookie, 5 min). Paste an example of the result you want. Examples steer format and tone more reliably than adjectives.
5. Pick the model and effort for the job (Rookie, 5 min). Use the model menu next to the send button. Low or medium effort for routine work, high for most things, max for the hardest thinking.
6. Get current answers with sources (Rookie, 5 min). For anything recent, let Claude search the web and check its citations before you reuse a fact.
7. Ask for a diagram or chart in the chat (Rookie, 4 min). Say “draw this as a diagram” or “chart this data” and Claude builds an interactive visual right in its answer.
8. Fix an answer with feedback, not a fresh start (Rookie, 4 min). Tell Claude exactly what to change and why. Two rounds of clear feedback beat ten new attempts.
9. Think out loud with voice mode (Rookie, 5 min). Tap the sound wave icon and talk a problem through. Claude can search and use your connected apps while you speak.
10. Use incognito chats for one-off questions (Rookie, 3 min). Click the ghost icon for a chat that is not saved to your history or to memory and is not used for training.
11. Share a chat with exactly the right people (Rookie, 4 min). Click Share and invite people by email. They see a view-only snapshot, not your files or later messages.
12. Make your usage last (Rookie, 5 min). Plan your ask, batch related questions, use projects for files you reuse, and save high effort for work that needs it.
13. Decide whether your chats help train models (Rookie, 3 min). Open Settings > Privacy and set “Help improve our AI models” the way you want. It applies to future chats.
14. Boss: your first real deliverable (Rookie, 12 min). Turn a real, messy input into something you would send today, using everything from this tier.

Apprentice

15. Write instructions Claude uses in every chat (Apprentice, 7 min). Put your role, audience and standing preferences in Settings > Your Claude > Instructions, once.
16. Teach Claude your voice and save it as a skill (Apprentice, 10 min). Show it your best writing, have it name the rules, edit them, then save the rules as a skill so every draft sounds like you.
17. Set up a project for one piece of work (Apprentice, 8 min). Put the files and instructions for one client, product or goal into a project once. Every chat inside it starts with that context.
18. Grow a project past the context limit (Apprentice, 6 min). Keep adding documents. When a project gets close to the limit, Claude switches to search mode and can hold up to ten times more.
19. Make memory work for you (Apprentice, 7 min). Claude saves topics as you chat. Review them in Settings, say “remember this” for key facts, and turn memory off for chats that should not count.
20. Bring your memory over from another AI (Apprentice, 5 min). Use Settings > Your Claude > Start import. Claude gives you a prompt to run in your old assistant, then you paste the result.
21. Make an artifact you can put in front of people (Apprentice, 8 min). Ask for a doc, deck, design or small tool. It opens beside the chat, you edit it with Claude or by hand, and you can share or export it.
22. Write living documents with Claude Docs (Apprentice, 7 min). Ask for a doc and Claude drafts it on screen, leaves comments on its choices, and lets you and your team edit together.
23. Turn notes into a deck with Claude Slides (Apprentice, 7 min). Give Claude your notes or report and ask for a deck. Edit slides directly, present from Claude, export to PowerPoint, PDF or Google Slides.
24. Mock up a one-pager or screen with Claude Design (Apprentice, 8 min). Ask Claude to design it and refine on the canvas. With your design system set up, every design uses your brand.
25. Get real Excel, Word and PDF files back (Apprentice, 7 min). Ask for a spreadsheet with working formulas, a Word report or a PDF, and download the file.
26. Use Research for questions that need many sources (Apprentice, 8 min). Turn on Research for questions that need five or more searches. Claude plans, searches in rounds and writes a cited report.
27. Connect Gmail, Calendar and Drive (Apprentice, 8 min). Connect your Google apps so Claude can search mail, find meeting times and read Drive files. It asks before sending or sharing anything.
28. Boss: make Claude know your job in 30 minutes (Apprentice, 30 min). Set up instructions, a voice skill, one project with real knowledge and memory, then produce a finished doc or deck from it.

Pro

29. Hand Claude a whole task, not a question (Pro, 8 min). Describe the outcome, the inputs and what done looks like. Claude plans the steps, uses its tools and hands back finished files.
30. Start a long task and walk away (Pro, 6 min). Cloud tasks keep running when you close your laptop. Tasks that touch your own files or apps need Claude Desktop open.
31. Schedule a task that runs every week (Pro, 8 min). Describe the task and the cadence in plain words. Claude proposes a name, schedule and instructions; you click Schedule.
32. Connect the apps you work in (Pro, 8 min). Add apps from the Connectors Directory. Claude can then read and act in them with exactly your permissions.
33. Add a plugin built for your role (Pro, 7 min). A plugin bundles skills, connectors and sub-agents for one job. Add one and you get a working setup on day one.
34. Write your own skill (Pro, 12 min). A skill is a folder with a skill.md file: a name, a description of when to use it, and clear instructions. One workflow per skill.
35. Let Claude work in your browser, safely (Pro, 10 min). Claude in Chrome can read, click and fill in forms on websites. Use it on trusted sites and watch for prompt injection.
36. Work inside Google Docs, Sheets and Slides (Pro, 7 min). Install Claude for Google Workspace and open it from Extensions > Claude. It reads the file you are in and edits it in place.
37. Connect Microsoft 365 at work (Pro, 8 min). After your admin sets it up, connect Microsoft 365 to search SharePoint, OneDrive, Outlook and Teams. Write tools need admin approval.
38. Ask your data for a dashboard (Pro, 8 min). Connect a data warehouse or an app like Salesforce, ask in plain language, and Claude writes the queries and builds the dashboard.
39. Claude Code for your first real change (Pro, 15 min). Open the Code tab in Claude Desktop, choose a project folder and ask for one small change. Claude shows a diff and asks before every edit.
40. Boss: hand over a whole weekly report (Pro, 30 min). Build a weekly report that gathers its own inputs from your apps, follows your skill, lands as a finished file and runs on a schedule.

Master

41. Give Claude a way to check its own work (Master, 8 min). Before you hand over a task, say how Claude can prove it is right: a test, a total that must match, a source to compare against.
42. Teach Claude Code your project with CLAUDE.md (Master, 10 min). Run /init to create CLAUDE.md, keep it lean, and after every correction ask Claude to add the lesson to it.
43. Plan first, then let it build (Master, 8 min). Press Shift+Tab into plan mode, refine the plan until it is right, then let Claude execute. If it goes sideways, re-plan.
44. Run several Claude sessions at once (Master, 10 min). Give each independent task its own session in its own git worktree, and say “use subagents” when one task should spread out.
45. Send coding tasks to the cloud and review the result (Master, 8 min). Start a cloud session from claude.ai/code, Claude Desktop, mobile or claude --cloud. Leave, and come back to a branch ready to review as a pull request.
46. Make rules automatic with hooks (Master, 10 min). A hook runs a command at a fixed point, for example formatting every file after Claude edits it. Use hooks for rules that must always happen.
47. Let Claude use apps on your screen (Master, 8 min). Claude uses connectors first, then a browser, and only then your screen. Turn on computer use for apps without any other way in.
48. Set up agents so mistakes stay small (Master, 10 min). Risk depends on what Claude can read and what it can do. Give write access only where you need it, and keep approval on for anything you cannot undo.
49. Give your team an agent in Slack with Claude Tag (Master, 8 min). Tag @Claude in a channel to hand it a task. Everyone sees the work, anyone can steer it, and it can follow up on its own.
50. Search your whole company with Ask Your Org (Master, 6 min). Use the “Ask Your Org” project to search across connected tools like Slack and Microsoft 365 in one place, with Anthropic-maintained instructions.
51. Know when Fable is worth it (Master, 6 min). Use Fable 5.1 for the hardest, highest-stakes work and Opus or Sonnet for the rest. On most plans it either costs credits or burns your limit faster.
52. Turn on automatic code review for your team (Master, 8 min). Code Review runs specialist agents on every pull request and posts verified findings as inline comments. It never blocks or approves.
53. Boss: three agents, one week, zero chaos (Master, 30 min). Run three real tasks in parallel for a week, each with a verification check, the narrowest tool access and a clear handoff.

Legendary

54. The master plan: put every piece where it belongs (Legendary, 15 min). Instructions for who you are, memory for facts, projects for context, skills for procedures, connectors for data and actions, plugins to bundle them, schedules for time.
55. Write skills once, use them in any agent (Legendary, 12 min). Agent Skills is an open standard: a folder with a SKILL.md. The same skill works in Claude and in dozens of other agents.
56. Connect Claude to any tool with a custom connector (Legendary, 15 min). Point Claude at a remote MCP server, your own or a vendor’s, and any tool becomes a connector. The server must be reachable from Anthropic’s cloud.
57. Build a plugin and list it in the Claude directory (Legendary, 20 min). Package MCP connectors and skills as a plugin, host it on GitHub, and submit it through the portal for review.
58. Build on the Claude Platform with your monthly credits (Legendary, 12 min). Claim your monthly API credits by linking a Claude Console organization, then build small apps and agents on the API.
59. Build your own agent with the Claude Agent SDK (Legendary, 20 min). The Agent SDK gives you Claude Code’s tools, agent loop and context management as a library you run in your own app.
60. Put Claude in your GitHub pull requests (Legendary, 12 min). Install the Claude GitHub app with /install-github-app, then tag @claude in issues and pull requests to have it fix, explain or update CLAUDE.md.
61. Learn from Anthropic’s own playbooks on GitHub (Legendary, 15 min). Work through Anthropic’s interactive prompt tutorial and the Claude Cookbooks. They show how Anthropic itself builds with Claude.
62. Teach it: run a one-hour Claude workshop (Legendary, 15 min). Teach five habits with live examples from your team’s real work, and leave everyone with one setup step done.
63. Your first Claude API app with prompt caching (Legendary, 14 min). Put the parts that never change first, add one cache_control field, and pay a fraction of the price for every repeat of that prefix.
64. Run big jobs at half price with Message Batches (Legendary, 12 min). When nobody is waiting for the answers, send the requests as one batch: all usage costs 50 percent less and most batches finish within an hour.
65. Boss: the 30-day Legendary quest (Legendary, 15 min). In 30 days, ship one system from each quest line: a shared skill, a custom connector or plugin, an agent, and a workshop.

Perplexity lessons

Perplexity is made by Perplexity (United States). The deck has 57 cards in five levels, from Rookie to Legendary. Every card was checked against current sources on 2026-11-05.

Rookie

1. Ask questions you can check (Rookie, 5 min). Perplexity shows where every part of an answer comes from. Use that: open the sources.
2. Write a search that gets a real answer (Rookie, 4 min). Say what you want, give the context, name the keywords and ask for a format.
3. Keep digging in one session (Rookie, 4 min). Ask follow-ups in the same session so Perplexity keeps the context and the sources.
4. Read the source labels (Rookie, 3 min). A shield icon marks Government, Academic or Trusted sites. No label is not a bad sign.
5. Pick where Perplexity looks (Rookie, 4 min). Choose the sources at the start: Web, Academic, Finance, your files, or none.
6. Ask questions about a PDF or a recording (Rookie, 5 min). Attach the file, ask a specific question, then check against the page or timestamp.
7. Get a full report with Deep Research (Rookie, 6 min). For a question worth a report, switch to Deep Research and let it work for a few minutes.
8. Leave it on Best, or pick a model (Rookie, 4 min). Best picks a model for each question. Pick one yourself only when you have a reason.
9. Share a session the right way (Rookie, 3 min). Sessions are private until you choose Share. Pick who can see it, and switch it back when done.
10. Search without leaving a trace (Rookie, 3 min). Incognito sessions skip memory and expire within 24 hours.
11. Decide what your data is used for (Rookie, 4 min). Turn off AI Data Usage if you do not want your searches used to improve Perplexity’s models.
12. Know what your plan gets you (Rookie, 4 min). Free covers everyday search. Pay when you need more Pro Search, models, files or Computer.
13. Boss: answer a real question you can defend (Rookie, 12 min). Take one real question from your week to a sourced answer you would put your name on.

Apprentice

14. Keep research together in Projects (Apprentice, 6 min). Put every session, file and instruction for one piece of work into a Project.
15. Give a Project instructions that stick (Apprentice, 6 min). Good project instructions say who the work is for, what to check and how to format the answer.
16. Research a competitor in 10 minutes (Apprentice, 10 min). Use a fixed set of questions and Deep Research to get a comparable profile every time.
17. Make memory work for you (Apprentice, 5 min). Let Perplexity remember preferences that help, and delete the ones that do not.
18. Tell Perplexity who you are (Apprentice, 4 min). A short profile with your role and preferences makes every answer fit you better.
19. Search data that is not on the open web (Apprentice, 6 min). Use Statista, Wiley, PitchBook, CB Insights and Midpage inside Perplexity, with citations.
20. Study with Learn Mode (Apprentice, 6 min). Learn Mode teaches with questions, hints, flashcards and quizzes instead of handing you the answer.
21. Get a document, sheet or deck back (Apprentice, 6 min). Name the file type you want, and Perplexity builds it with its research and citations inside.
22. Make images that fit your work (Apprentice, 5 min). Describe subject, style, colours, mood and setting. Pick a default image model in Settings if you have a favourite.
23. Ask questions across your Drive or Dropbox (Apprentice, 6 min). Connect your cloud storage so answers can use your own documents, with citations.
24. Browse with Comet, Perplexity’s browser (Apprentice, 6 min). Comet puts Perplexity next to every page: ask about it, summarise it, or tidy your tabs in plain language.
25. Boss: build your research base (Apprentice, 15 min). Set up one Project that makes Perplexity know your work, then prove it with a real deliverable.

Pro

26. Hand your first task to Computer (Pro, 8 min). Search answers questions. Computer does the work: give it a whole task and a clear finish line.
27. Pick the right effort level (Pro, 5 min). Light for routine work, Standard for everyday briefs, High and Ultra only when the problem needs it.
28. Know what a task will cost (Pro, 5 min). 100 credits is 1 dollar. A quick fact is tens of credits, a dashboard can be thousands.
29. Install a plugin for a whole job (Pro, 6 min). A plugin bundles the connectors and skills for one job, so Computer gets everything at once.
30. Edit a file by pointing at it (Pro, 5 min). Select the part you want changed, describe the change, and submit all edits at once.
31. Put recurring work on a schedule (Pro, 7 min). Describe the job and the cadence in plain words, confirm, and manage it on the Automations page.
32. Send Computer a task by email (Pro, 4 min). Email or forward a task to computer@perplexity.com and get the result back in the same thread.
33. Work with Computer inside Slack (Pro, 5 min). Mention Computer in a thread and it reads the thread, does the work and posts the result there.
34. Build and publish a small website (Pro, 8 min). Ask Computer to build it, publish to a pplx.app link, and set who can see it.
35. Ask three models and see where they disagree (Pro, 6 min). For decisions you will act on, let several models answer and read where they agree and differ.
36. Turn your routine into Comet shortcuts (Pro, 5 min). Save a prompt you use every week as a /shortcut and run it anywhere in Comet.
37. Boss: hand off a weekly job end to end (Pro, 15 min). Take one real weekly task off your plate with Computer, a plugin and an automation.

Master

38. Write a skill Computer uses on its own (Master, 10 min). A skill is a reusable method. Its description decides when Computer uses it, so write that first.
39. Connect an API without exposing the key (Master, 7 min). Give Computer your API key through the secure form, never in the chat.
40. Add your own MCP connector (Master, 8 min). Point Perplexity at a remote MCP server over HTTPS and its tools become available to Computer.
41. Let Brain carry your work context (Master, 6 min). Brain builds a working model of your projects, people and files, with every memory linked to its source.
42. Work on your own files with Personal Computer (Master, 7 min). The desktop app gives Computer your local files, apps and a keyboard shortcut from anywhere.
43. Keep sensitive work on your own hardware (Master, 7 min). Portable Computer runs the agent and model on your device, and asks before using the cloud.
44. Keep agents on a short leash (Master, 7 min). Give agents read access first, approvals for actions, and the fewest connectors they need.
45. Connect private Mac data with a local MCP server (Master, 10 min). Install the PerplexityXPC helper, add the server’s run command under Add custom, and keep sensitive files on your own machine.
46. Build a skill toolkit Computer combines on its own (Master, 9 min). Install Perplexity’s skills from Discover, add your own, and write sharp descriptions so Computer can chain several skills in one task.
47. Boss: an agent that knows your work (Master, 20 min). Build a skill, connect a tool safely and run a real task that uses both, with brakes on.

Legendary

48. Make your first Agent API call (Legendary, 12 min). The Agent API answers from the model’s knowledge unless you add the web_search tool.
49. Prompt an agent loop, not a chatbot (Legendary, 10 min). Put rules in instructions, the question in input, and hard limits in parameters, not prose.
50. Give your coding agent Perplexity search (Legendary, 8 min). Add Perplexity’s hosted MCP server to Claude Code, Cursor or VS Code for current, cited answers while you build.
51. Search the web from your terminal (Legendary, 6 min). The pplx CLI returns Search API results as JSON for scripts, pipelines and coding agents.
52. Package and share your own plugin (Legendary, 12 min). Bundle your skills and MCP servers into one .zip so a team installs your method in one step.
53. Classify and route with the Decisions API (Legendary, 10 min). When software needs a yes, a choice or a score, ask for probabilities instead of parsing text.
54. Keep Agent API cost and behaviour under control (Legendary, 14 min). Start from an effort level, cap the loop with max_steps, pin a version where you need stable behaviour, and log usage.cost.total_cost on every response.
55. Feed your own model with the Search API (Legendary, 12 min). The Search API returns ranked web results as JSON, with no written answer, so your own model or agent decides what to do with them.
56. Search your own documents with the Embeddings API (Legendary, 12 min). Turn your documents into vectors once, store them, and find the closest passages to any question by cosine similarity.
57. Final boss: ship a research product (Legendary, 30 min). Build something other people use that answers with fresh, cited facts, and run it like a product.

Chat & assistants

• ChatGPT by OpenAI (United States). The all-round assistant for writing, thinking and quick analysis. 5 lesson cards.
• Claude by Anthropic (United States). Careful writing and long documents, in your own voice. 4 lesson cards.
• Gemini by Google (United States). Google’s assistant, built into Gmail, Docs, Sheets and Drive. 45 lesson cards.
• Grok by SpaceXSI, formerly xAI (United States). An assistant with live access to what people are posting on X. Three preview tasks.
• Meta AI by Meta (United States). A free assistant inside WhatsApp, Instagram and Messenger. Three preview tasks.
• Copilot by Microsoft (United States). AI across Windows, Edge and Microsoft 365. Three preview tasks.
• Vibe by Mistral AI (France). Mistral AI’s EU-hosted agent for work and coding, formerly Le Chat. Three preview tasks.
• DeepSeek by DeepSeek (China). Strong reasoning from open models, free to use. Three preview tasks.
• Qwen by Alibaba (China). Alibaba’s assistant and open models, strong in many languages. Three preview tasks.
• Kimi by Moonshot AI (China). Built for very long documents and deep research. Three preview tasks.
• Doubao by ByteDance (China). ByteDance’s everyday assistant, hugely popular in China. Three preview tasks.
• ERNIE by Baidu (China). Baidu’s assistant, built around Chinese search. Three preview tasks.
• GLM by Zhipu AI (China). Open models from Zhipu AI with strong coding and agent skills. Three preview tasks.

Research

• Perplexity by Perplexity (United States). Answers with sources you can check. 3 lesson cards.
• Gemini Notebook (formerly NotebookLM) by Google (United States). Turns your own documents into answers, summaries and podcasts. Three preview tasks.
• DeepL by DeepL (Germany). Translation that keeps your tone. Three preview tasks.
• Elicit by Elicit (United States). Finds and summarises research papers for evidence-based answers. Three preview tasks.

Images & design

• Midjourney by Midjourney (United States). Striking, stylish images from a few words. Three preview tasks.
• FLUX by Black Forest Labs (Germany). Photorealistic images from the team behind Stable Diffusion. Three preview tasks.
• Canva by Canva (Australia). Design anything, with AI doing the first draft. Three preview tasks.
• Figma by Figma (United States). Design and prototype, with AI help built in. Three preview tasks.
• Adobe Firefly by Adobe (United States). Images and edits built for commercial use, inside Adobe apps. Three preview tasks.
• Ideogram by Ideogram (Canada). Images with clear, correct text, like posters and logos. Three preview tasks.

Video & voice

• Runway by Runway (United States). Professional AI video generation and editing. Three preview tasks.
• Kling by Kuaishou (China). Realistic AI video with smooth, natural motion. Three preview tasks.
• Hailuo by MiniMax (China). Fast, expressive video from text or an image. Three preview tasks.
• ElevenLabs by ElevenLabs (United States). Natural voiceovers, dubbing and sound. Three preview tasks.
• Suno by Suno (United States). Full songs with vocals from a short description. Three preview tasks.
• Luma by Luma AI (United States). Smooth, cinematic video from text or a single image. Three preview tasks.
• Pika by Pika (United States). Playful short videos and effects for social. Three preview tasks.
• HeyGen by HeyGen (United States). Presenter videos with AI avatars, translated into other languages. Three preview tasks.
• Synthesia by Synthesia (United Kingdom). Training and explainer videos with AI presenters, for companies. Three preview tasks.
• Descript by Descript (United States). Edit video and podcasts by editing the text transcript. Three preview tasks.

Work & automation

• Notion by Notion (United States). Ask your workspace and stop writing follow-ups. Three preview tasks.
• Zapier by Zapier (United States). Connect your apps and let AI do the busywork. Three preview tasks.
• Make by Make (Czechia). Visual automations for bigger workflows. Three preview tasks.
• n8n by n8n (Germany). Open-source automation with AI agents you control. Three preview tasks.
• Cohere by Cohere (Canada). Enterprise AI that works with your company’s own data. Three preview tasks.
• Gamma by Gamma (United States). Presentations, docs and simple sites from a prompt. Three preview tasks.
• Grammarly by Grammarly (United States). Fixes and polishes your writing wherever you type. Three preview tasks.
• Otter by Otter.ai (United States). Meeting notes and transcripts, written for you as you talk. Three preview tasks.
• Fireflies by Fireflies.ai (United States). A meeting assistant that records, transcribes and searches your calls. Three preview tasks.

Build & code

• GitHub Copilot by GitHub (United States). An AI pair programmer inside your editor. Three preview tasks.
• Cursor by Anysphere (United States). The AI-first code editor. Three preview tasks.
• v0 by Vercel (United States). Build web apps and interfaces from a prompt. Three preview tasks.
• Replit by Replit (United States). Describe an app and ship it from the browser. Three preview tasks.
• Bolt by StackBlitz (United States). Prompt, run and deploy full-stack web apps in the browser. Three preview tasks.
• Lovable by Lovable (Sweden). Build full web apps just by chatting. Three preview tasks.