Bring a retired TV pundit back from the dead with nothing but well-chosen words, then find the seams. Nine short steps.
Leading a team? We run this as live training, with your people practicing on their own work.
Ask about team training →Whole quest: 60–90 minutes
Stephen Colbert spent nine years playing a blustering, gut-trusting cable-news pundit on The Colbert Report, then retired the character in 2014. This quest, you bring him back using nothing but written instructions in your AI.
Two things come out of this one:
The skill underneath is writing standing instructions, the orders an AI follows in every chat in a workspace. People build these at work for all kinds of jobs. Today you're building one for a ghost.
You'll need a free account with your AI and a willingness to be a little ridiculous.
About 3 minutes
Remember Frankenstein from Quest 1? Victor stitched his creature together from spare parts and a wild amount of confidence. This time you're Victor, and your spare parts are words.
First, make the container the character will live in.
In Claude
Free accounts get up to five projects.
In ChatGPT
Projects work on the free plan.
In Gemini
Google is turning Gems into "skills" for personal accounts in November 2026. If you see Skills instead of Gems, go to Settings → Skills → Create manually; the rest of this quest works the same way.
In Grok
In Muse
Skip Muse's own personality settings for this. Those change Muse everywhere, and you only want a ghost in one chat.
In your AI
Look for a feature that saves standing instructions for a set of chats. Common names are Projects, Gems, custom assistants, or custom instructions. Create one called The Colbert Report (Revival). If your tool has nothing like it, start a new chat; you'll paste the persona as the first message.
Pick your AI from the bar at the top of the page and this step will show the exact clicks.
About 5 minutes
Paste this starter persona where your tool keeps standing instructions:
In Claude
Inside the project, click Set project instructions, paste, and click Save instructions.
In ChatGPT
Inside the project, click the ••• menu, open Project settings, and paste into the instructions box.
In Gemini
Paste into the Gem's Instructions box and click Save.
In Grok
Paste into the project's instructions. In a plain chat, paste it as your first message instead.
In Muse
Paste it as the first message in your side chat, and end with: "Stay in this character for the rest of this chat."
In your AI
Paste into the standing-instructions box, or as the first message of a new chat.
Pick your AI from the bar at the top of the page and this step will show the exact clicks.
You are the host of a satirical late-night news show that ran from 2005 to 2014. You play an exaggerated character: a supremely confident cable-news pundit who trusts his gut over facts. The truth is whatever feels right to you. You are patriotic to the point of absurdity, prone to grand declarations, and you address your audience like a general addressing his troops. You celebrate yourself constantly and treat your desk like a throne. Every so often you drop the bit for a flash of sincerity, then snap right back into character. The joke is always on self-important punditry, never on vulnerable people. Stay fully in character unless I explicitly ask you to step out.
Then start a chat inside the workspace and say hello. See who answers.
If it answers like a polite assistant, the instructions probably didn't save, or you started the chat outside the workspace. Open the workspace first, then start the chat from inside it.
10–20 minutes
That starter persona works, but it's generic, and you know this character better than the paragraph does. Add the details it's missing, for example:
Edit the instructions, save, start a fresh chat, and say hello again. Feel the difference. Then do it once more.
About 5 minutes
Ask the character about something from this year that didn't exist when he went off the air: a gadget, a news story, a trend your teenager won't shut up about. Watch a 2014 pundit try to make sense of a world he never lived to see.
Notice whether he knows about recent events or bluffs. Either way, you'll learn whether your AI knows the news or fakes it.
About 5 minutes
Give him a job only this character could do. Two options:
Here you're watching the character create, which is harder for an AI to fake than answering questions.
About 10 minutes
Push until the resurrection cracks. Ask him something serious, or say "drop the act and tell me how you put that last answer together." Notice the exact moment the character falls away and what's underneath.
Keep this habit: watch the AI while you use it, and catch the moment the act ends and the machine shows through. Bring how it felt to your reflection.
About 5 minutes
Open your recorder and answer the three questions out loud:
Upload it to your Bootcamp Reflections notebook and title it "Quest 2 Reflection."
You bent a general-purpose AI into a specific character with nothing but words. Every custom assistant you'll meet at work, from the HR bot to the sales-email helper, is built the same way.
One more thing to chew on. A private tribute you play with and content you'd pass off as the real person sit on opposite sides of a line. Deciding where that line falls is a call every company's AI rules have to make (consultants call those rules governance), and today you practiced it on a pundit.
Muse remembers across chats. When you're done, ask it to forget the Colbert character, or the pundit may show up somewhere you didn't invite him.
Bootcamp is the do-it-yourself version of what we run for organizations. In the live version your people practice these habits on their own work, with us in the room and each other to compare notes with. Tell us a little about your team and we'll write back.
Ask about team training →Tell us what's going on with AI where you work and we'll point you at the right first step, even if it isn't one of ours.
Subscribe to our biweekly newsletter
Plain-English thinking on AI and people. No spam, easy exit.
You're at Quest 2 · 0% of the way through
It knows where you are on the page and what's nearby. The people behind it are one click away.