It always goes like this:

1. You make a good prompt.

2. AI isn’t good enough, so you correct it.

3. AI finally makes it well. But then the chat is long gone.

4. So you start a new chat for the same task. The cycle repeats.

Because you must teach AI skills: a repeatable task you teach it once.

As the Chinese proverb goes:

Give AI a fish, and you feed it for a day.

Teach AI to fish, and you feed it for a lifetime.

Every AI you know has the skill feature:

So you can transfer all of your skills from one AI to another.

Now, how do we make them? Here are the best three ways:

1. Recording your screen to create an AI skill.

2. Using a skill to create skills from scratch.

3. Turning a (good) chat into an AI skill.

You can stop the newsletter here, create skills yourself, and you’d be good to go.’

But if you give me another 7 minutes, I will cover:

• for rookies → the screenshots of the entire setup (and the mistakes to avoid, especially when recording your screen).

• for advanced users → how to download my 3,457 skills library covering every single job (that is not coding).

Share

This newsletter is free because people like you share it with people they love.

1. How to teach AI skills.

• If it’s multiple steps on your computer → record your screen.

• If it’s something you know by heart → use a skill creator from scratch.

• If it’s from a conversation you just had → ask AI to turn the chat into a skill.

I. Record your screen.

II. The skill maker.

I made a skill to make skills.

Claude also made one. Let’s start with it:

I love to use this simple prompt to make skills:

I want to create a skill that [task] for [goal] with [success criteria].

An example:

I want to create a skill [that reviews CRE leases and LOIs against my standard deal positions] for [deciding what to accept, negotiate, or send to counsel] with [every deviation flagged by severity and suggested redlines].

Now, if you want to use my own home-cooked skills to make skills, you can download it here: https://www.dropbox.com/t/HFlo6HVbVHYmiExj.

If the Dropbox download does not work, open it in incognito (or on your personal computer; sometimes companies prevent it for security reasons).

And then you upload a new skill like this:

III. A chat into a skill.

This one is fairly simple. On any chat, search for this feature:

A quick reminder that a skill is not a scheduled task.

For a regular task (like every morning), you need to go here:

I also love Grok Bot for these kinds of tasks. I wrote about it in ‘why I hate agents’.

Share

2. How to download my skills.

I am lazy. I hate writing very long prompts every time I open Claude.

So I built a skill library over the years.

Then I started training companies on ‘How to AI’.

So the library is now 3,457 skills sorted by jobs.

Download them right here:

Read more

She’s a retired pharmacist.

So she has no reason to care about AI.

But being my (Jewish) mom, she wants to understand what her son is teaching to 900,000 readers. And I want to finally answer the question, “Who’s Claude?”.

(Claude is a French name, I’m French; it checks out)

For my mom, AI is not Claude. It’s not even ChatGPT. AI is just Gemini (because she has an Android and they are pretty aggressive with Gemini).

Claude is this small to the rest of the world:

So for this newsletter, I will teach you Claude the same way I taught my mom Claude. And by the end, you will know what matters, your wrong assumptions about it, and even use it the way so-called Claude pros do.

But before we start, I’d love you to send this article to your mom.

It’s free, and it will bring her in the AI world:

Share

This guide works with every AI, ChatGPT, Gemini, or Grok. Soon, they will be so similar it will be like picking up sneakers.

1. So that’s Google, right?

No Mom, Claude (and AI) is not Google.

It’s tempting to think it is because you search for something and it magically answers. But it’s closer to your phone trying to predict the next word.

This is how AI works.

Let’s take the sentence “This is how AI finds the next […]”

Claude goes through 4 steps to determine […]:

My mom then asked, “But how does it know which word?”, so I had to explain very simply how AI gets trained. It read so many books and internet discussions that it got a good feeling for how sentences are made.

Was it done legally and ethically? Meh.

Anthropic trained on millions of single-unit books, and then burned them:

But so did ChatGPT when they trained their AI. And now new AIs are distilling other AIs (it means they are training themselves on other AI answers).

If we must learn something from history, it’s that technology can’t be stopped. It can be contained, regulated, safely deployed - but never stopped.

So I am happy my mom retired.

Because she never enjoyed tech so much. And the world is dividing in two: either you’re extremely good at handling frontier technology (like AI), or you’re so likable people keep coming back to your shop (that’s my mom at the pharmacy).

But the “likable = job” isn’t so obvious.

For example, we used to have porter as a job:

I’m sure they were very nice and likable, like my mom - but then we invented wheels on luggage, and the roads became easier to navigate.

Technology replaced likable people.

If your mom needs to get better on the technology side, with AI, let’s continue the newsletter.

Subscribe now

2. Her first prompt.

Prompting is easy, Mom, you just need to structure it this way:

I promise, you won’t need more than this.

Let’s take an example: my mom had to pay for a notary to plan her succession. It was a technical and legal question (that is costly to ask!!).

It involves her property, her marital status, the fact that she has kids (we’re four), and tons of new French laws no one reads and stays on top of.

So we started with this prompt:

Now the truth is, my mom can’t type this.

She’s a very slow typer.

So I showed her Wispr Flow, a dictation tool. We simply talked like she would to her notary.

At some point she got scared: “What if I don’t say enough?”. But that’s Claude’s job. That’s why we added, “If necessary, ask me clarifying questions first.”

I love Claude for this reason. It creates an easy form:

10 minutes in, and my mom got every answer the notary told her… without the cost! She couldn’t believe it.

“It answers so well! Maybe we should share it with the notary?”

I’m not done, Mom.

Searching and finding answers is one step.

Let’s create the right strategy & write the contract:

Share

3. More than answers.

From that point onward, my mom understood what makes AI different than Google.

You can literally write the contract:

1. A complete strategy document.

2. The pre-filled contract a notary can review.

3. All of this, almost for free, within <10 minutes.

I told my mom, “Imagine if you started the conversation with the notary having these documents first. Would it be the same price?”

I believe not.

That’s how you expand someone’s brain about AI: they finally feel how it widens their options. They are now capable.

But my mom needs more. She’s a tough cookie.

“How can we be sure of this ‘Claude’? What is its source?”

You’re right. It’s called dealing with hallucinations.

Subscribe now

4. Dealing with hallucinations.

‘Hallucination’ is what we call AI making things up.

We say that “The AI hallucinated [something].”

And it’s annoying.

In a recent paper, they found that if you show the AI a fancy-looking chart, it thinks it’s real. The AI doesn't check whether the numbers are true. It just trusts that the picture looks official.

So naturally, my mom got scared.

But you can force Claude or other AI to cite their sources and pick official ones only. Just add:

It looks like this for my mom:

And it looks like this inside Claude:

And this is true with every “but AI is bad” comments → people are one-sided because they had one bad experience or they heard one bad story.

Yes, hallucinations exist.

Yes, you can fix it.

Yes, if you use a public Claude and you share private data, it has some risks.

Yes, if you use a Business Claude, they are SOC 2 compliant, so it can’t leak that data. The same way Microsoft or Google works.

Once she felt confident, she asked me the final question:

Subscribe now

5. Should I open Claude?

No, you don’t, Mom. You’re retired.

Enjoy your grandchildren.

That being said, you will most likely use a product from Claude without even noticing. Because it’s so easy to build anything with Claude.

Let’s build one together: remember the strategy and the contract we wrote for her specific situation? What if we turned it into a website so any mom in her situation could have the results?

“We can do this?” Yes, we can:

6. How I use Claude.

Less and less like I taught my mom.

More and more with Claude Code instead.

Now I’m not technical. I don’t know how to code. But I’ve been spinning up live websites almost every day for any use case with this simple prompt:

Read more

For example, agriculture and the wonders of the Green Revolution brought cheap + convenient + abundant food.

And we are now so fat we must fight obesity.

AI makes thinking just as cheap, convenient, and abundant.

So what will happen to our intelligence?

If we need to go to the gym today to work out and burn our cheap and abundant calories, do we now need to go to mental gyms for our brains?

I wrote this guide to answer it.

For my own sanity, in a post-AI world.

Share

Sharing is caring. Share this with someone who is scared of getting dumber because of AI. That person is both right and wrong. I’ll cover both.

1. Eyeglasses Do Not Make You Blind.

Technology isn’t automatically making you a bad human.

1. Eyeglasses do not make you blind.

2. Writing made memory less important, but we are accumulating knowledge like never before.

3. Calculators made mathematicians better. Not worse.

4. GPS makes traveling safe. Imagine a plane without GPS.

5. AI can answer, sure, but also teach.

The variable is not convenience itself.

But some convenience, once eliminated, also wipes out repetition, attention, error correction, retrieval, senses, and learning.

The things that make you a functioning human.

Here are some examples:

An ultra-processed diet caused people to consume 508 ± 106 additional kcal/day and gain about 0.9 kg in two weeks, while the same participants lost roughly 0.9 kg on the unprocessed diet.

And we struggle to expend these new calories because we move less. So energy ( = calories) became easier to get while becoming less necessary to expend. [1]

The cognitive analogy is stronger with GPS and AI.

A small three-year study found that heavier GPS use was linked to faster decline in spatial memory. [2] The more you use a GPS, the less you can navigate by yourself. Ouch.

What about AI? In a 2025 randomized experiment with 1,000+ high-school math students, ChatGPT improved performance by 48%. But when the AI was removed, those students scored 17% worse than the control group.

The lesson is simple: the task got easier faster than the student got better.

But when ChatGPT was only a tutor, performance jumped 127% without the same drop afterward. [3]

So AI makes it easy to answer, so fast, you don’t have time to learn. But if you use AI to learn, you outperform everyone else.

Medicine too. After experienced doctors began using AI to detect polyps during colonoscopies, their detection rate without AI fell from 28.4% to 22.4%. [4]

This problem existed before AI. Airline pilots using highly automated cockpits kept their basic flying skills fairly well. But they became worse at tasks the automation usually handled: knowing their position without a map, planning the next navigation step, and diagnosing instrument failures. Regulators therefore recommend that pilots regularly practice flying manually. [5]

Memory shows the same pattern. When people know information will remain available on a computer, they remember less of the information itself and more about where to find it. Taking photos can have a similar effect: people who photographed museum objects later remembered those objects and their locations less well than people who simply looked at them. [6]

Our senses can change in the same way. What we repeatedly practice noticing becomes easier to detect and describe. Hunter-gatherers in Malaysia could name smells about as easily as colors, unlike a closely related farming population living in a similar environment. In another study, five months on a low-salt diet made people more sensitive to salt and reduced how salty they preferred their food. What we repeatedly use, we become better at sensing. [7]

Cheap calories let us consume more while moving less. Cheap cognition lets us produce better answers while thinking less. In both cases, the immediate result can improve while the underlying human ability gets less practice.

But there is a twist:

Offloading is not the same as losing a skill.

It becomes deskilling when we stop practicing a skill we still need, and perform worse when the tool is removed.

The point of this newsletter is to make sure AI benefits us.

And how.

I mean, I literally called this newsletter “How to AI”.

I am just like you. I do not want to get dumber because of AI.

Subscribe now

Upgrade today to join our Community.

2. The “use-it-or-lose-it” test

To test whether a technology is causing loss of a skill, ask four questions.

First: what did you have to do before?

You might have had to walk, remember a phone number, build a mental map, spot cancer, solve a calculation, read the weather, navigate by stars, cook from raw ingredients, find a partner through friends, or fly a plane manually.

Second: and now what does technology do for you?

Cars replace walking. Processed food replaces preparation. Search engines store information. Cameras store memories. GPS chooses routes. Dating apps find and filter people. Autopilots control flight paths.

How many phone numbers do you remember? Mine. And my mom.

And now AI comes for our thinking.

Third: do you still practice?

A tool can make us better if it helps us practice. But if it does the task before we do, you won’t practice, so you end up losing the skill.

Fourth: what if the tool disappears?

Let’s say the tool disappears.

If you’re now worse than before = problem.

And with AI, it’s already happening.

I saw this Reddit post the other day, and I had chills:

I assign a profile essay. Have for 30 years. Student interviews a person, records it, writes a paper based off the recording. Citations are timestamps. Should be pretty easy. Has been easy pre-AI.

Now, students cannot do it.

I get so many requests for alternative assignments. Students claim they don’t know anyone. Claim anxiety. Want a list of questions from me.

Even after the interview:

Student: “So, what do I put in the paper?”

Teacher: “Information from the interview.”

“Yeah, but what parts?”

“You have to decide that.”

“But how?”

“Start with what you found interesting.”

“How do I know if something’s interesting?”

“You’re asking me how you find something interesting?”

Kids can’t know if something is interesting because AI did not tell them if it was. Is this it? The end?

Now wait. Not everything is lost.

We can still use AI to our benefit.

Share How to AI

Most of this newsletter stays free because people like you share it with people they love & respect. Be one of them.

3. Zero-Effort Thinking & Meat Proxy.

The danger is not cheap intelligence.

It is zero-effort thinking.

If you indulge, not only will your brain and intelligence suffer, but you’ll perform workslop. San Francisco invented a new word for you: you are a meat proxy.

So how do we get to choose when to AI and when not to AI?

When the goal is output → automate aggressively.

When the goal is learning → make the user (you?) predict, draft, recall, explain, decide, and correct mistakes before showing the final answer.

In practice: instead of asking AI what strategy to choose, tell the AI the many strategies you thought of → if there is any you didn’t think of → explore each one's pros and cons → be the pilot, not the assisted passenger seat.

I gave it a thought. I searched. I shared where my brain is going. What I want to validate and where I might need to take a step back. AI is a partner, not the single source of truth.

Because no, you can’t escape AI.

AI keeps getting better & cheaper, faster.

You think AI is getting more expensive because of Claude-Fable-5 or companies that keep spending more and more? You’re thinking about it wrong.

AI is thinking for us, more and more.

And that thinking is getting cheaper and cheaper. More and more abundant. So convenient it’s almost impossible to avoid it.

Just like we go to the gym to work out, we might have to work out with our brains to still be sharp. Thinking right can’t be outsourced entirely.

A gym puts physical effort back into a world that no longer requires much of it. We may soon need cognitive gyms that deliberately restore memory, calculation, navigation, writing, judgment, diagnosis, social interaction, and problem solving after AI makes much of that effort unnecessary.

So I built a mental gym for staying relevant as humans.

First, copy the secret password here:

Read more

But what if I had zero money?

I tried the free tier of hundreds of AIs to answer the question.

Share

Before you start, take a second to share this newsletter with anyone who’s struggling to pay for AI.

#1. The best free chatbot.

ChatGPT:

I need the best chatbot to do three things well.

1. It must be smart enough.

2. It must have very few limits.

3. It must have enough useful features.

First, for the smartest, I like to use this Intelligence Index:

Now the problem is that most of these models are not available for free.

If you only focus on the free ones, the list is very small:

1. Claude Opus & Fable are only part of the paid plan.

2. Claude free model is Sonnet 5, and it’s not a good model.

3. Gemini free model is 3.6-Flash, far from the top 20 models.

4. Grok is great, but not free. Keep an eye on it for the future.

5. The Chinese ones (Kimi, DeepSeek, Qwen) are cheap if you use them to code (you pay per token). But I want the option of having no money at all.

That’s why ChatGPT is the best for a 100% free account.

But especially if you use ChatGPT like this:

1. You must download the app. I believe every LLM is better on its native app (vs. the browser). They optimize everything for it: connectors, skills, models…

2. Inside the app, go to “ChatGPT Work” and choose the model “5.6-Terra-High”.

3. Once you've tapped out Terra, you can always use the smaller model, Luna, in High settings. I believe this one is free forever without limits.

But how does it compare to the paid Claude?

Actually, very good.

I needed to build a quick spreadsheet for a client, and made this prompt:

I sell software for $170/year.

I pay a team $50,000 per month, and I want to give them 20% of the revenue (shared across the team). I also spend ads, so far very little ($1000/day), and I only have to spend $50 to sell a $170/year software. The churn is about 70% after year 1, and 30% from year 1 to year 2.

How much money I can scale and spend on ads, while still giving bonuses to my team, and being able to pay myself a salary of $50,000/mo.

If you click on Plugins, you can select Spreadsheets, and ChatGPT will do much better ones. You even have some templates you can preselect:

And you can connect many apps, for free, inside ChatGPT.

Just follow these steps:

So ChatGPT is the best chatbot for free. It is connected to most apps and has an excellent free model (Terra-High in Work, or Luna-High always free).

And a bonus: you can also make images:

Chatbot is 99% of your AI needs.

But what about the rest?

Share

#2. The best free [task].

ChatGPT covers most of your (free) needs.

But here’s a quick list of the rest:

1. Gamma for slides: 400 starter AI credits, but they do not refill. You can create AI presentations, use built-in themes, publish/share on web and export. You can earn more credits via referrals.

2. Wispr Flow for dictation: Desktop: 2,000 words/week soft cap, 5,000 hard cap. iOS: 1,000 soft / 1,500 hard. Dictation works across apps, 100+ languages, Mac/Windows/iOS/Android. Caps reset weekly.

3. Shortcut AI for Excel: if your life is to make Excel (otherwise ChatGPT free is more than enough). 20 AI credits/week, recurring. Includes web + desktop app + Excel and Google Sheets plugins.

4. Seedance for videos: Dreamina gives free daily credits and access to Seedance (the best AI model for videos). Currently 225 free credits/tokens per day, resetting daily.

5. Google Vids for videos: another good one, although I prefer Seedance/Dreamina. Anyone with a Google account gets 10 Veo 3.1 video generations/month.

6. Grammarly for grammar fixes: Unlimited basic spelling/grammar correction, but they are very (very) annoying with the pop-up to make you pay.

7. NotebookLM for learning: One of the strongest free tools for research over your own sources, with notebooks, source-grounded chat, and tons of other stuff. I think the new name is Gemini Notebook, though.

8. Cursor for coding: it’s a bit weird to code for free, and ChatGPT Codex is already good too, but Cursor has a hobby plan: free, no card, with limited Agent requests + limited Tab completions.

9. ElevenLabs for text-to-speech: Free text-to-speech/voice AI. One of the obvious category leaders. Obviously the free plan is limited.

10. Granola for meeting notes: Free meeting-notes tier. Particularly good if you want an invisible meeting bot. I prefer.

11. Canva for designs: Huge permanent free design product. Some AI/Magic Studio capabilities are available for free but heavily limited.

And if you are vibecoding with ChatGPT Codex (or Cursor), here’s my favorite combo, free until you start hitting some volume limits:

1. Vercel: So that your website is on the internet, deployed for free.

2. GitHub: So that the code of your website/project is stored somewhere.

3. Supabase: So that your website has a sign-in option (so imagine you want to have a unique experience per user; there is a login option).

4. Resend: So that you can send instructions by email to people who connect to your website (works well with logins, for example). 3,000 transactional emails/month for free.

I very often start to vibecode anything by saying:

You are already connected to my Github, Vercel, Supabase & Resend.

I want to build [project] for [goal + success criteria].

Make sure that [rules, max 3]. Ask me clarifying questions first.

This was the list of the best AI tools for free.

Now here’s how much I pay for AI, and for which tools:

Read more

And AI changes so fast that techniques from before don’t work anymore. LLMs (like ChatGPT or Claude) changed so much it’s outdated.

For example, I used to tell everyone to add “Take a deep breath and work on this step by step” to their prompt. Because science said so:

But this paper is from September 2023 and talks about the model PaLM 2-L from Google. You now have access to Claude & ChatGPT models that are 10x more capable.

So how can you truly know how to prompt the latest models?

You need to read new and trusted academic papers on prompt engineering. Hundreds of them. Per day.

But you must be crazy to read 100+ new AI papers every day.

Well, I am crazy.

By the end of this newsletter, you will know what the science says about better prompts (in simple English, I promise; no weird Claud-isms), in this order:

1. Never end your prompt with “right?”.

2. Forget about the “step by step”.

3. Do not trust confidence.

4. Stick to 3 rules max.

5. Goals > Examples.

Sounds like a good deal? Cool.

Two things before we start:

1. Save this guide. It’s long, so block 15 min on your calendar.

2. Send it to anyone who is searching for better prompts.

Share

This newsletter grows from your shares. It’s my weekly north star.

1. Never end with “right?”.

A Cornell Tech researcher tested 45 different AIs with one-word changes.

1. “[X] is the better choice, right?” → new models disagree with you.

2. “[X] is the better choice, maybe?” → all 45 models agree with you more.

So asking the AI’s opinion is most definitely not a good idea: it will be biased by how you ask the question more than by its actual intuition.

Bad prompt:

“I’m deciding what to do about housing.

Buying is the better choice, maybe?”

Good prompt:

“I’m deciding what to do about housing.

Compare buying and renting for my situation.”

Just like when you ask a friend, don’t make them answer the way you want them to answer. Keep it neutral. Avoid yes/no questions.

Subscribe now

2. Forget the “step-by-step”.

IBM ran an impressive 430,738 evaluations on 8 prompting techniques.

The most famous “Let’s think step by step” LOST to asking normally.

The winning combo is just your question + your role in a few words (“…as a reliability engineer”).

Bad prompt:

“[Your question]

Gather information, devise a plan, answer step by step.”

Good prompt:

“[Your question]

. . . as a reliability engineer.”

Worth noting. The paper used the role "reliability engineer" for every topic it tested: medicine, physics, and math. So is it the only role you should prompt, or does choosing a good role make your prompt better?

Maybe I should start writing academic papers…

Share

3. Do not trust confidence.

Microsoft’s CTO audited 2.6 million references (papers) at the world’s top AI conferences.

1 in 4 ‘NeurIPS’ 2025 papers — papers that PASSED expert peer review — has a hallucinated citation. The reviewers missed it, and they even scored the papers slightly HIGHER.

If professional reviewers can’t catch AI-hallucinated papers, you won’t either. You must open every link:

Now the real question is who in their right mind will open 210 sources.

I feel the same as you. I wish to trust AI research more.

Subscribe now

4. Stick to 3 rules max.

Meta tested GPT-5.5, Claude Opus, Gemini Pro and 12 others AI on prompts with 1 to 12 rules. An example of 3 rules would be:

Exactly 3 paragraphs. Under 150 words. No emojis.

At 8 rules, models only get each individual rule right 41% of the time. But it only gets 8 rules at once 5.7% of the time… Not good.

12 of the 15 models can’t reliably hold more than 3 rules.

Bad prompt:

Write a LinkedIn post about our launch. Exactly 3 paragraphs. Under 150 words. No emojis. Include “AI-native”, “workflow” and “ship”. Don’t mention competitors. End with a question. Grade-6 reading level. Match my voice sample below.

Good prompt:

First run: Write a LinkedIn post about our launch. Must include: "AI-native", "workflow", "ship". No emojis. End with a question.

Second run (after the draft): Check your draft against each of these requirements ONE AT A TIME, then revise: 3 paragraphs, under 150 words, grade-6 reading level, no competitor mentions.

The more I learn about prompt engineering, the more I understand it’s all about giving the right goal. A clear one. It’s far more effective than giving examples:

Share

Sharing is caring. And it’s free.

5. Goals > Examples.

Researchers found examples in your prompt are sabotaging modern models.

Mistral’s model scored 74% with the standard examples prompt. They deleted the examples, and then boom → 83.8% score.

Bad prompt:

You are a world-class newsletter strategist.

Example 1: [a solved case, written out]

Example 2: [a solved case, written out]

Find a plan, and answer step by step: How can I improve my newsletter?

Good prompt:

My newsletter open rate fell from [X]% to [Y]% over three months.

I send one issue every Tuesday. I did not change the format, the send time, or the subject. What are the most likely causes?

Ask me for the data you need before you answer.

Specifying a goal is the ultimate key to a good prompt.

6. The One Prompt

I made a prompt template that covers everything we just covered.

I will then show you how to create a quick keyboard shortcut (because you don’t want to copy & paste it every time you need it).

So first, copy and paste this prompt:

Read more

I have 900,000 subscribers on both Linkedin & Substack, so naturally people ask me questions (please keep asking btw):

And yet, it’s extremely simple. Just type:

Prompt: “/setup-cowork start”

And Claude will literally set up Claude Cowork, for you.

You can stop the newsletter here, use the skill /setup-cowork, and follow the steps, and you’d be good to go.'

But if you give me another 12 minutes, I will cover:

• for rookies → the screenshots of the entire setup (and the mistakes to avoid, especially for cloning your voice),

• for advanced users → how I use Claude these days (spoiler: less and less with Cowork).

Share

This newsletter is free because people like you share it to people they love.

1. How to set up Cowork.

Once you type the skill /setup-cowork + start, you need to complete the entire setup form it generates.

Instead of writing how I answered, I will share screenshots.

Gmail/Outlook might block you from seeing the entire email because it’s too many screenshots. So you should open the article here.

If this was confusing, here’s a quick video:

Now I told you about the mistake not to make when setting up Claude.

Claude Cowork wants to build your “writing voice”.

I wrote a lot about mirroring your voice with Claude. And you can do it to some degree. But I deeply believe writing is thinking. AI can help you search and form an opinion, be the best 24/7 thinking partner you can have, but you won’t be a Linkedin/Substack star by writing with AI. And you certainly won’t be a loved coworker/boss if your reports are AI-slop.

AI is very good at many things. But some things are better kept unique & flavorful. And I think writing (especially publicly) is one.

Share

2. How I use Claude.

Less and less with Claude Cowork.

More and more with Claude Code instead.

Now I’m not technical. I don’t know how to code. But I’ve been spinning up live websites almost every day for any use case with this simple prompt:

Read more

But what the fuck is an agent?

No one knows:

That’s why, even though I have over 800,000 people reading my newsletter on How to AI, I never mention “agents”.

But today, this insanity stops.

You & I both will know what an agent is, and we will build our first one together.

I’ll assume the word “agent” makes you think of James Bond, so we really start from zero.

Our first agent will do something that takes too much of your time: keeping track of AI news (only the one you care about). A simple, safe & useful agent.

For the pros, I added a last section called “How pros use agents” at the end.

Two things before we start:

1. Save this guide. Block 10 min this week to create your first agent.

2. Send it to anyone who is scared by the word “agent”. I am also scared btw.

Share

PS: This newsletter grows from your shares. And I keep hitting 1,000+ shares! It’s my weekly north star. Sharing is free & helps me stay laser focused on mastering AI.

1. Chatbot or Agent?

You’ve been using AI chatbots your whole life: you ask, it answers.

But an AI agent is doing a task, start to finish:

And it’s important to understand the difference.

Because AI agents are getting cheaper than… humans.

It’s a scary image, I know.

But until very recently, I couldn’t care less about agents.

Too clunky to build, too much work to maintain them.

I still think most companies should focus on AI adoption instead of trying to build agents. That’s why my consulting firm - GPC - trains employees before building any agents.

Send me a message if your company has over 100 employees in the US, especially in the manufacturing industry.

And yet, I kept trying to make agents. Easy, useful ones, connected to my apps.

Like technically, when I use Claude Code or Cowork, they are agents.

But you’re still using it as a chatbot. You ask. It answers.

These were my only “agents”. Until I tried this one:

Subscribe now

2. The easiest way to build a good agent.

I will stop you right there.

I am not paid to write about this company. They just have a good, useful tool to make agents. Especially for us, people who don’t code (or don’t want to).

It’s called Grok Bot, a tool made by xAI*.

*SpaceX bought xAI, the AI company of X, which makes Grok models.

On the left: You see my “bots,” or call them agents.

In the middle: When it finishes work, it updates you here. You can also interrupt it anytime to give a new task, or how you feel about its execution.

On the right: You see the agent computer (yes, it has its own computer) and its routine (how often should it perform a task, or many tasks?).

Before I show you exactly how to access it & install it (just go to x.ai/bot if you’re in a hurry), here’s an example of an agent I built:

To connect your apps to Grok Bot:

How to install Grok Bot:

1. Go to x.ai/bot.

2. You must have a Cursor Ultra/Teams or Grok Heavy X account.

3. The least expensive is Cursor Teams ($40/mo).

Once done, we can create our first agent together.

I think it’s much simpler for you if I explain with screenshots:

My exact prompt: “Daily digest of AI news in my industry (or very close to it).”

My exact prompt:

“What I care about, as a daily digest:
- the top 3 news of AI by joint virality (The Rundown, Google News, what pops on Reddit, cross-check them)
- the top 3 news in the creator economy
- the top 3 news in the consulting business
- read ruben.substack.com and update me on a recent "how to" guide”

Now this was one agent. A simple one, useful, fast to make.

But here’s my recipe to create an advanced agent (it’s a copy & paste prompt):

Read more

But you can fix it in 2 words for 80% of your chats.

Prompt: “Use ASD-STE100.”

Let me show you the difference:

You can stop the newsletter here and use the trick on every AI you know (ChatGPT, Gemini, Grok… but also coding platforms like Cursor, OpenClaw, Hermes…).

But if you give me another 10 minutes, I will cover:

• What ASD-STE100 is, and why it is so effective with AI.

• My plug-and-play ASD-STE100 Claude skill (free to download).

• How can I write with AI to 800,000 people (if it’s a slop machine).

Share

This newsletter is free because people like you share it to people they love.

1. The ASD-STE100 magic.

ASD-STE100 is simplified English for instructions.

So when you ask AI to adhere to ASD-STE100, it works.

AI knows enough about the ASD-STE100 to reply with simplified English.

Now we could go the extra mile and build a Claude skill with the rigorous ASD-STE100 list. But who is crazy enough to build such skill?

Me.

And I built a skill to get Claude to truly talk like ASD-STE100 instructions.

1. Download it for free at: https://drive.google.com/file/d/1ZYmir1Ian-duOJip7rSSWwkdRQXRCYh0/view?usp=sharing.

2. Then go to Claude to install it (it also works on ChatGPT):

OK now should you use the prompt or the skill /ste all of the time?

No.

It’s awesome to use, but falls short on some tasks:

Subscribe now

2. My copy & paste research prompt before every piece of text.

This section is to copy my shortcuts, and see it live.

I am recording myself writing a newsletter (to 885,000 readers) and the Linkedin post that goes with it (I have 900,000 followers there).

The goal?

Never sounding like an AI (even when you didn’t intend to).

So first, copy and paste this prompt on Claude or ChatGPT:

Read more

Because more of your brain is used to process images, not text, to understand the world.

So when I’m writing on Linkedin, most of my work is actually making images.

And that’s what got me here:

People forget what they read. They remember what they saw.

Now I know what you’re thinking. This is yet another AI guide, and you’re gonna follow it and make infographics that look like everyone else’s, and only I, the author, get good results with it.

Yes, this is an AI guide. But also, yes, if you follow it, you’re going to get good results that look like yours. Money-back guarantee.

Two things before we start:

1. Save this guide for later. It’s long, so you need a moment to digest it. Book 15 minutes in your calendar and follow each step as we go through it.

2. Send it to the guy at the office who only posts awesome walls of text that deserve the best infographics.

Share

This newsletter is free because people like you share it to people they love.

1. Everything is code, even you.

If you really, really deeply think about it, everything is code.

That’s why AI labs go so crazy about code, because code is how to make a website, but also images, videos, Excel. Probably all of your job can be coded, and that’s why we’re seeing the world changing so fast with AI.

Now, I’m not here to say that you’re going to disappear and nothing matters, but just that you can make cute, cool, digestible infographics with code without even knowing how to code.

1. Go to Claude.

2. Make sure to use the home tab. The code tab overcomplicates infographics.

3. On the bottom right, select the model, Opus or Fable. They will do a better job, but Fable is expensive, so I usually use Opus.

4. Upload one or multiple images to help Claude figure out the style you need and upload them. Guess what? If you can process images better than text, it’s the same for LLMs.

5. Now copy and paste this prompt:

6. Once you’re happy with the style preview, paste in the information you want in your infographic (or you can ask Claude to search for you before making it).

7. Generate your image.

If you have a hard time following these instructions, here is the step-by-step process with screenshots:

Upload your images and paste the prompt from above.

Now generate.

When you’re done, you should get something like this:

Claude calls this a “design system.”

Once you’ve checked that these elements are to your liking, ask the model to make your infographic with your information using the system it just gave you.

Notice how it’s double-checking itself so thoroughly? That’s how you know we’ve got a good prompt.

The big upside of coding an infographic is that it is never wrong because it’s like the difference between drawing a table and an Excel spreadsheet.

An HTML image is precise and doesn’t have the mushy feel of some AI images. Check how easy it is to edit it:

Here’s what it gave me:

Now, let’s say you just made a super cool infographic and you want to remake it in the same style but with different information.

Go back to your original prompt, and simply press “Edit” to change the topic:

Press this button and rewrite what information you want.

I went from an infographic about information retention to one about caffeine consumption. And this is still going to work.

This works exceptionally well when your reference image is of high quality.

My favorite place for that is Pinterest, for pretty much everything. What I do is that I make a query with one or two words max. Press enter. I click on the image I like the most, and then Pinterest sends you down the rabbit hole of images. It gets better and better the more you click on new images.

For web design, you should try Awwwards, Godly, CSS Design Awards, Dribbble, or Refero. Copy an image, upload it to Claude, that’s your reference.

Try it. It’s very fun.

Share

This newsletter works because people like you share it with people they love.

2. ChatGPT is still good at something.

This is the technique everybody knows to make AI images.

It’s called diffusion.

You type a description, and AI paints you pixels.

And the #1 leader on the market is clearly ChatGPT:

Now, it’s pretty easy to prompt it. You just:

1. Upload an image you love.

2. Paste your information or ask ChatGPT to search.

3. Generate the image, in any ratio, and edit it over & over again.

Here’s the step-by-step:

If you are reading this in Gmail, it doesn’t support my super-cool-text-heavy instructions. You don’t want your guide cut off, so get the full experience here:

Open the full article.

This tells the system right away that you’re looking for an image, not text.

Now upload your references and tell it what you want.

I chose some visually complex images to show you just how accurate diffusion models can be.

But as I said, you can edit it.

When should you use ChatGPT (+diffusion)?

• visual-heavy infographic

• you don’t need to change data much

• you prefer visual control over data control

When should you use Claude (+HTML)?

• data-heavy infographic

• you need to change numbers quite a lot

• you prefer 0 mistakes over artistic output

But wouldn’t it be better to have…both?

Subscribe now

Upgrade your subscription to access my community.

3. You want the bread & the butter.

Me too buddy, I am super French.

Remember that fix I promised you?

Well, here it is → Don’t choose. Use both.

1. Let diffusion (so ChatGPT) build the background.

2. And let code (so Claude) build the charts and numbers.

Here’s how:

Open a new chat in ChatGPT.

Again, click the “plus” and “Create image.”

Write your background image prompt and generate.

Pro tip: remember to ask it to leave empty space for text and charts.

Now open Claude, upload the image ChatGPT just made, and ask it to use the picture you just generated as the background for an infographic.

I made sure to clarify that I wanted an HTML infographic. That’s important.

If you made a system using the method I showed you, use it.

Now you’re combining static artwork (diffusion) with live elements (code).

Make infographics (even) faster.

You can teach Claude once how you love infographics, as a skill.

Brand colors. Typography. References. Even your favorite screenshots.

But before I teach you how, I want you to know the problem with this strategy: it’s very repetitive. So it’s perfect if you must stay consistent, but not if you want a wide range of infographic styles.

Here’s the step-by-step:

Bring back the design system we created earlier. It’s not a good time to be a skimmer. Go back to the first section of the newsletter.

Click on the “Design system” file that ends in .md to open a preview.

Then click the down arrow on the top right, and click “Download as MD,” just like this:

Now open a new chat and upload the design system you just downloaded.

Like so:

With that uploaded, we can start creating a skill.

Write or paste: /skill-creator

Now, copy and paste this prompt into the chat, then generate:

Once it’s done generating the skill, you’ll get the option to save the skill.

Now you can use this skill whenever you need to make an infographic, and your prompts can be painfully simple.

Just look.

This is the result:

Share

My future AI will spare anyone sharing this newsletter for free.

You’re good, but your team is not.

Teaching one person AI is not the same as having a hundred people adopt it inside a company. And every week with my team, we talk to CEOs and chiefs of staff who get AI, and just don’t understand why it takes so long for the company to benefit from it. They just never cared to see how a normal person uses AI (spoiler alert: they don’t, or they pretend they do).

If you feel like I’m calling you out because you have an AI roadmap that no one follows, or you have no AI roadmap and you’re wondering how your business is going to go in the next couple of years, I’d love to talk to you.

You can click on the big button to message me.

I promise, it works.

But if you have fewer than 100 employees, I have something else for you.

I built a community on Circle to help you to connect with me once a week on a call.

I won’t be able to build your entire roadmap or make workshops for your entire company, but I’m sure I can still help you set you up for success.

So first, click on the link if you want to be part of the Circle:

Read more

I started waaaay before you, back in March 2023, when this was the design:

Since then, I have been begging you to switch.

First in December:

These two newsletters are completely outdated for mastering Claude.

If I had to get you up to speed very fast, here are my 27 (non-obvious) tips to get you to adoption faster.

Share

This newsletter works because people like you share it with people they love.

#1. Stop it with your Claude ‘Projects’.

You have a Claude Project to write your Linkedin posts.

Something that looks like this:

You loaded 40 files into a Project, and now every answer is a remix of your own documents. Claude answers FROM your files instead of thinking.

That’s why your post all sounds the same.

Project is awesome if you want to write a contract or standard procedures. But it’s not as good for creative work.

The rule: use Projects for repeated tasks with fixed material (client reports, weekly formats). Use a normal, empty chat every time you want a new idea.

#2. Fable-5-High first. Then Opus-5-High.

You frame your entire problem on your first prompt; it’s the most important one. So that’s when you want to use the most powerful Claude model (Fable-5).

Then, switch to Opus 5 in High to continue the conversation without spending extra dollars. Here’s a step-by-step:

#3. Claude can’t technically make images.

Claude can’t make images like ChatGPT.

But it can generate an HTML (and then you export it as an image).

Prompt: Code an HTML-like infographic like the one I attached, but about [topic].

Bonus: the text is always spelled correctly in HTML, which image generators still can’t guarantee. Most graphics in this newsletter are made this way.

#4. Don’t reply “no, that’s wrong”. Edit instead.

When you correct Claude, the wrong answer stays in the conversation. Claude keeps working around it, and you pay for it on every message after.

Because Claude always reads eeeeveeerything in your conversation before answering back. That’s why long chats can be a burden.

Instead, do this:

#5. Better yapping than typing.

When you type, you organize your thoughts, and you delete context without noticing. When you talk, you don’t. You share everything, even the messy parts.

Your “prompt engineering” will be 100x better.

1. Install Wispr Flow (what I use for dictation) or use the one from Claude.

2. Talk for 10 minutes straight: the goal, the constraints, what you tried, what you hate, your contradictions. Don’t clean anything. Keep the mistakes even.

3. End with: “These were messy voicenotes. Ask me clarifying questions if you didn’t get something as a form.”

I write this newsletter like this.

*If you want the recording of this video, I am sharing the link after the paywall at the end of this newsletter.

#6. Turn off the Connectors you’re not using.

When you click on the “+”, you can see Connectors to connect your favorite apps (like Gmail, Slack, Granola…).

The problem is that every active Connector gets loaded into every single message you send, used or not. So you pay extra as “token usage”. Just turn them off unless you actively use them.

#7. When Claude gets dumb, start a new chat.

Long conversations get more expensive and less sharp with every message.

Because Claude re-reads everything. And then it tries to make sense of too many things at once. If you are contradicting, Claude now has to deal with it.

It’s useful sometimes, but I’ve seen some people's chats, and I can safely say: be careful with your suuuuuper long chats.

Just start a new one.

#8. Ask Claude to ask YOU questions.

You don’t know exactly what you’re asking for.

Don’t worry. Me neither.

So add this magic line to any prompt: “Before answering, use the AskUserQuestion form to get more context from me if necessary.”

Claude will show you tappable multiple-choice questions (the AskUserQuestion tool, the most underrated feature in the app). And the questions are good.

AI is all about sharing enough context.

#9. Code is better than Cowork at everything.

Claude Code is better than Claude Cowork at everything.

But it looks scary & made for devs. That’s the only problem.

I do think that in the future, both will merge and be approachable for anyone. But until then, you should at least try once:

1. Install Claude. Go to the top left “Code” tab.

2. Click on the dictation tool and ask to vibecode whatever you want.

3. Sky is the limit. Like try: Build a website to track my daily frisbee sessions.

It does not matter if you master Claude Code. You just need to try it once and know it exists. It will expand your understanding of what’s possible.

You can always hire someone to use Claude Code for you later on.

#10. Cowork is for many Claude, without coding.

Everyone sells Cowork as “Claude on your computer”.

I mean it is, it can create files and folders inside your computer. But the real feature is to spawn agents. Multiple Claudes attacking one problem in parallel.

So give it big tasks, not small ones: “Prepare the full client onboarding: the deck, the welcome email, and the checklist.” Then watch them split the work.

#11. How to set up Cowork with a skill.

There’s a skill (setup-cowork) that interviews you and configures your Claude: preferences, styles, workflows. And it’s made by Anthropic’s team.

#12. Send a screenshot instead of a description.

In Claude Code and Claude Design, always start with an image: a competitor’s page, a dashboard you like, a photo of a napkin sketch.

To make a design, it’s better to see something than to read words.

PS: It also works on the normal Claude to make HTML designs.

#13. Build your own mini-apps.

You can create artifacts with Claude. It’s mini-apps.

Try this prompt: “Build me an artifact that tracks [my habits / my client pipeline / my reading list], and make it save my data between sessions.”

#14. Invite Claude in your mini-app.

So the same idea as #13, but with Claude inside your app.

For example, this prompt: “Make it so there is a Claude coach inside I can talk to and it has the context of this artifact.”

Then publish the artifact and share the link: a proposal analyzer for your clients, a quiz for a new hire, a tone-checker for your team. They don’t need to use Claude. They use Claude through your mini-app.

#15. Stay under 150 seats.

Premium seats (at $100/seat/month) come with usage limits high enough that your team stops thinking about them. But as soon as you cross 150+ seats, you pay per usage no matter what, and it gets very expensive.

#16. Combine Connectors together.

My favorite combo when writing a big email. I connect:

• Slack = what was said on our shared threads

• Granola = what was decided in the meeting

• Gmail = what was promised in writing before

And now I go to Claude with this prompt: “Using Slack, Granola and Gmail, draft 3 different emails with 3 different tones about […].”

#17. Share what you hate.

Adjectives do almost nothing, like when you prompt “make it nicer/punchier”.

It’s much better to prompt: “Never write like this: [paste].”

It draws the exact line to not cross.

#18. Open files in Google Drive.

I rarely open Claude documents on my computer.

Instead I click here:

#19. Vibecoding won’t make you rich.

Vibecoding a software won’t make you rich. You need so much more things than code to “be rich”.

But next time you have an idea, don’t write a briefing document. Build a rough version in one afternoon and send THAT to your technical team.

This is why vibecoding is cool: you can talk to engineers and designers with an actual (half-baked) website you made quickly. This is useful.

#20. Skills = repeated tasks. Not creative work.

Skills are the best AND the worst.

The best for “do this exact task this exact way”: reports, formats, recurring workflows.

The worst for creativity: every loaded skill bloats your context and narrows what Claude explores. Which is the opposite of what you want for a creative task.

Sometimes I lean even more into “anti-context-bloating,” and I open an incognito mode inside Claude to make sure it has zero context cluttering (I wrote cluttering myself without AI, I know, crazy).

#21. No one uses the “Research” mode.

But it’s good.

It’s not a slower search. It plans, reads dozens of sources, and comes back with a structured report you’d normally pay an analyst for.

Toggle “Research” in the chat bar, ask your decision question (”Should we price at X? What does the market do?”), and go get a coffee.

#22. Make your data interactive.

Paste your numbers (or a csv.) and ask: “Make this an interactive chart.”

It will make it interactive, right inside your chat. It’s very nice.

#23. Don’t write like an AI.

I built a skill that audits any text to make it passable for AI detectors.

A full guide here:

#24. Ignore “loop engineering” and whatever technique comes out next month.

Every month there’s a new technique with a super-serious name, and you feel behind. You shouldn’t care less. A technique matters once it becomes invisible.

Like for example, Chain-of-Thought prompting gave us reasoning models. You don’t need to understand CoT, you just know that the AI thinking before answering gives you better answers.

Focus on doing your job better, not on falling in love with the technology.

#25. Don’t switch to ChatGPT.

ChatGPT is good. Anyone selling you a 10x difference is selling something.

But I still prefer Claude. My reasons are narrow, though. Claude writes and reasons slightly better, and the Premium company tier is a better deal (tip #15).

The only scenario where I’d reconsider: 5,000+ employees.

At that scale, an open-source setup starts to make sense.

#26. If you use Google, Claude is just as safe.

Back in 2022, when AI was just ChatGPT, people assumed it was terrible for privacy. It was true back then, but today’s AI (like Claude or ChatGPT) is just as safe for your data as Google.

They have business plans and are SOC-2 compliant. You can read this document (link) and sign up for a team account if you have at least 2 seats.

PS: Fable-5 does have a data retention problem. The other models are fine.

#27. An AI solved a math problem that stayed open for 80 years because someone asked.

Some say:

AI is data.

Data is the past.

Creativity looks forward.

So AI can’t be creative.

It’s both true and false.

Hardcore mathematical problems that had remained unsolved for 80 years were solved simply because someone asked the AI, with the right context.

And that was not even a one-off.

Someone else did it again on X:

Share

Share this with someone who says AI is not creative.

Watch me apply the 27 tips.

And get $200 of free AI credits. You just need to become a paid subscriber.

I will share both the link to my next live session & the code to the free AI credits after this paywall:

Read more

The CEO of Substack just announced their new AI detector:

You can also detect AI slop instantly.

What’s AI slop?

It’s when you receive this from your colleague:

This budget isn’t a number. It’s a statement of intent. Because at the end of the day, a budget doesn’t tell you what you can afford — it tells you […].

You don’t need a degree to detect it.

If you and I can detect AI so easily, clearly Substack bot can detect it too, right?

Right??!

Well, it’s more complex than that.

But at the end of this newsletter (it’s 8 minutes, don’t be dramatic), you will master:

• How Substack AI detector works.

• How to forever not sound like an AI.

• How I write my newsletter with AI, live (and send it to 800,000+ readers).

Subscribe now

I. How Substack AI detector works.

Substack uses Pangram to detect AI.

I have zero affiliation with them, but I would honestly love to: I tried to game their tool for 3 hours, spending $34 of Claude Code credits to try to bypass it.

And I couldn’t. Pangram is really good at detecting AI.

But I still found a way to bypass it.

Here’s how I tried to game it (and succeeded at the end):

Sometimes it worked, sometimes it didn’t.

You can download it here if you want to try it yourself: https://www.dropbox.com/t/fORTSF4grrSCzrMN (the password is RUBEN-HOWTOAI).

Then you go here to Claude to upload the zip. file:

I made the simplest video on how to install it:

Now, how does my skill perform against Pangram?

Let’s test it.

So at this point, you think we can’t game Pangram.

I also thought so.

Then I removed the — and replaced it with a “:” (check my circle on the screenshot). Here’s the new result:

So yes, detecting AI is a bit of a scam.

We can’t (100%) predict AI with bots.

Even one of the creators of ChatGPT said it multiple times:

AI detectors are just pattern-matching machines. They learned what “AI text” looks like. And what “human text” looks like.

But humans can write robotic text. AI can write natural text. The two overlap. A lot.

Short text gives the detector almost nothing to work with. Imagine if I ask “AI or not?” with a one-liner. The overlap between “human” and “AI” is basically a circle.

What if I prompt AI for a certain style? What is then considered AI-writing is no longer AI-writing. And I can change the style infinitely.

See the difference:

But if the best detection bots on earth can’t catch AI text... how did you catch your coworker in 3 seconds?

It’s a lazy use of AI, with incorrect prompting, and it leads to the typical AI-writing style like:

• Over-simplification: “Most people […]”

• Meta-commentary: “Here’s the thing:”

• Negative parallelism: “It’s not X, it’s Y.”

• Abuse of adverb: “[…] quietly runs […]”

• Words you’d never use: “delve” or “tapestry” or “foundational”

AI is everywhere, and used poorly.

I am mostly known for my Linkedin account (and its 900,000 followers).

Well, a study estimated that 53.7% of long-form LinkedIn posts (100+ words) in 2025 were AI-generated. And you can be sure people used it lazily.

Chris, the CEO of Substack, a brilliant leader I met last May in SF (I wish I had a pic, but then you’d see that Chris is 1 meter bigger than me), does not want Substack to become Linkedin.

I am with you, Chris. We must chase & kill AI slop the same way we chase & kill bots on social platforms. My entire Linkedin comment section is just AI slop. I can’t find a single human inside it. There is a better way to do it. To write. To think. And yes, still with a bit of AI.

So here’s how to forever not sound like an AI on Substack:

Share

This newsletter is free because people like you share it for free to people they love.

II. How to forever not sound like an AI.

I am “Mr. AI,” and I hate AI writing.

Let’s say someone from my team sent me some AI slop.

Now their work just became my work. I have to decode it, guess what they meant, rewrite it, or send it back.

There’s a word for this: workslop.

You don’t want to do workslop. So you need to recognize AI slop first.

1. The worst AI style (with examples)

The constructions that give it away fastest, worst first.

Structure & rhythm giveaways

• Low burstiness. uniform 15–20-word sentences, rectangular paragraphs. Force a spread: a ≤6-word sentence and a 25+-word one; uneven paragraphs; one single-line paragraph, max.

• Fractal summaries. previews and recaps at every level (”In this section we’ll…” / “…as we’ve seen”). Delete all.

• Signposted conclusion. “In conclusion / Overall,” + restatement + uplift (”Despite the challenges, the future is bright”). Delete; end on the last concrete point.

• Pep-talk ending. “As we move forward, embracing X will be key.” Delete on sight.

• Prompt echo. “This essay will explore…” Delete.

• Listicle in a trenchcoat. “The first reason is… The second reason is…” Merge into flowing argument.

• Uniform staccato. “X is A. X is B. X is C.” Combine into one sentence with a list, or vary the frames.

Punctuation & formatting

• Em dashes — the most famous tell. AI: 20+ per piece; humans: 2–3. Target ≤1.

• Bold-first bullets. “Security: …”. Almost no human does this unprompted. Yes I just did it here. Yes I did ask Claude to summarize my findings.

• Emoji bullets. ✅ 🧠 🔹 decorative →. Strip them.

• Title Case Headings / colon-split titles. “The Power of X: Why Y Works”. Use sentence case.

• Oxford comma 100% of the time. dropping it occasionally in casual registers reads human.

• Markdown residue. **, ##, [text](url) in contexts that don’t render markdown.

4. Content & voice

• No concrete imagery, if the first three sentences evoke nothing visible, inject a thing, place, number, or name.

• Proper-noun avoidance, “a client”, “a tool”, “a city” → name it. (AI invented-character names cluster on Emily/Sarah.)

• Uniform positivity, everything upbeat and certain (measured: certainty +111–152%, positive emotion +69–133% vs human). Let something be annoying or unresolved.

• Both-sidesing, every claim auto-balanced by its counterpoint. Commit.

• Suspiciously tidy anecdotes, stories that serve the argument with perfect efficiency. Real stories have tangents.

• Register scrubbing, no contractions, no slang. Restore the ones the voice would use.

2. The worst AI words (the entire list)

Tier 1 — never use this

Verbs: delve · leverage · underscore · harness · foster · navigate (figurative) · utilize · facilitate · streamline · bolster · illuminate · showcase · embark · elevate · empower · unleash · unlock (figurative) · uncover · optimize · garner · resonate · revolutionize · shed light on · synthesize · elucidate · transcend · reimagine · intertwine · entwine · grapple with · espouse · exemplify · underpin

Nouns: tapestry · landscape (figurative) · realm · ecosystem (figurative) · paradigm · synergy · testament · beacon · journey (figurative) · interplay · intricacies · symphony (figurative) · kaleidoscope · tempest · whimsy · quest (figurative) · roadmap (figurative) · endeavor · myriad · plethora · advancements · trajectory (figurative)

Adjectives/adverbs: pivotal · crucial · seamless(ly) · robust · vibrant · intricate · meticulous(ly) · nuanced · cutting-edge · transformative · game-changing · groundbreaking · unparalleled · invaluable · multifaceted · commendable · indelible · poignant · profound(ly) · relentless(ly) · tireless(ly) · unwavering · unyielding · timeless · ever-evolving · fast-paced

Stock phrases: in today’s fast-paced world/landscape · it’s important to note · it is worth noting · plays a pivotal/crucial role in · stands as a testament to · rich tapestry/history/heritage · navigate the complexities of · in conclusion · in summary · overall, (as an opener) · ultimately, (as a conclusion opener) · at its core · that being said · a key takeaway · paving the way for · valuable insights (into) · deeper understanding of · shed light on · when it comes to · not only… but also · here’s the kicker/the thing/the best part · I hope this email finds you well · look no further · dive/deep-dive into · let’s explore/unpack/break down · furthermore · moreover · additionally (sentence-initial)

Narrative clichés: couldn’t help but feel/wonder · heart pounding · a sense of X washed over · found solace in · the human spirit · from that day (on/forward) · little did I/we know · a stark reminder · a cautionary tale · knew that (he/she/they) had to · felt a (newfound) sense of purpose · what lay ahead · turn of events · thick with (tension) · stumbled upon · nestled (in/between) · bustling · enigmatic · captivating · glimpse into

Tier 2 — allowed alone

comprehensive · significant(ly) · essential · critical · key (adjective) · dynamic · innovative · powerful · notable/notably · vital · vast · rich (figurative) · deep/deeper (figurative) · explore · enhance · ensure · foster · highlight · reveal · engage · embrace (figurative) · insights · perspective · framework · approach · strategy · challenges · opportunities · potential · impact(ful) · quietly/quiet (figurative “quiet confidence”) · genuinely · truly · remarkably · arguably · generally speaking · typically · thought-provoking · well-being · resilience · perseverance · dedication · commitment to · high-quality · step-by-step · sustainable/sustainability

Swap table:

3. The other things that make you say “that’s AI”

Never do this:

• Leaked scaffolding, “Certainly! Here’s…”, “I hope this helps”, “let me know if…”

• Self-reference, “as an AI language model”, knowledge-cutoff notes

• Placeholder text, “[insert example]”

• utm_source=chatgpt.com in URLs

• Hallucinated-looking citations

• “Best regards” sign-offs in non-email contexts

Be careful with this:

• Performative, text that narrates its own helpfulness (”I hope this clarifies things!”).

• One-point dilution, the same idea restated in new clothes across a paragraph.

• Curly quotes, pasted into plain-text contexts.

• Semicolons, used where a period would do (model-dependent, both directions).

• Emily / Sarah and other clustered default names for invented people.

What NOT to do when fixing it:

• Don’t swap every word with weird ones, we can tell.

• Don’t scatter random typos, errors must read as casualness, not carelessness, and only where the register tolerates them.

• Don’t scrub personality along with the tells, a flat, tell-free text is still AI-shaped.

• Don’t invent real-sounding facts, stats, or quotes.

• Don’t shrink every long sentence, humans write long sentences; they just don’t write only 18-word ones.

Human markers to add (the good list):

• Contractions

• a number with texture ($43, 11 months, 4:30am, v2)

• a named thing (brand, tool, street, person)

• a parenthetical aside with attitude

• “I think” / “honestly” / “to be fair” used once

• a sentence starting with And, But, or Because

• one single-sentence paragraph

• a mild complaint or unresolved edge

• an irrelevant-but-true detail in an anecdote

• a dropped Oxford comma (casual registers)

• a question the reader was genuinely asking

• uneven list items · a plain “is” where AI would write “serves as”.

Share

III. How I write with AI, live.

I send a weekly newsletter to 849,273 readers.

This section is to copy my exact writing process.

I started a Circle community to host monthly lives on how I use AI.

The goal?

Writing a newsletter from start to finish with you, so you can see how I use AI and you can ask me questions during the live.

All of the prompts, skills, and processes will stay forever on the Circle.

So first, confirm you want to be part of the Circle:

Read more

They announced:

• A new mega-app for ChatGPT.

• Codex, their coding tool, is now inside ChatGPT.

• A new mode inside the ChatGPT app, called ChatGPT Work.

• The new flagship model (5.6) with three levels of intelligence: Sol (very smart), Terra, and Luna (very cheap).

This entire newsletter will answer one question:

Does ChatGPT still matter?

Yes, it matters.

1. OpenAI is one of the two leading AI labs in the world.

2. ChatGPT will always matter, with its 5 billion monthly visits:

But does it matter enough to switch from Claude?

Harder question. Let’s find out.

1. How to access the new ChatGPT.

You need to:

1. Uninstall the old ChatGPT app (if you had it before).

2. Download it (again, I know) at chatgpt.com/download.

3. You must have a paid account (at least $20/month).

It will look like this once you open it:

Now that you have access to the new ChatGPT, it’s time for you to know the truth.

2. It’s so confusing, I’m lost.

These are the 30 combinations of the new ChatGPT.

ChatGPT Work? Codex? Which model? Which effort? Good luck.

Let me answer as simply as possible.

The first distinction is Work vs. Codex:

• ChatGPT Work is to work on anything but code. Like Claude Cowork, it spawns many ChatGPT (agents) to answer your queries.

• ChatGPT Codex is for developers. It’s the same, but with a design and an interface made for developers.

So you want to stay on ChatGPT Work.

Now, the second one is the different models:

• You have 5.5 (old one) and 5.6 (the new one).

• 5.6 comes in 3 flavors of intelligence…

  • Luna: cheap but dumb.

  • Terra: the cost/smartness balance.

  • Sol: the super smart super expensive one.

If that was it, that would be so easy… but no… you also have different efforts.

It’s quite obvious that the more effort, the smarter the model, but the more expensive it is. What’s not obvious at all is which one is better between

• 5.6 Sol (the smartest one) on Light effort?

• or 5.6 Luna (the dumbest one) on Extra High effort?

• and also, WHY DO WE HAVE TO PICK??? I can’t believe it.

So I tried to make a graph to explain it better… but it’s also a bit confusing. I’m sorry. I don’t work at OpenAI or Anthropic. I also wish it were easier.

I’ll save you some thinking time:

• The best expensive one: Sol-Ultra, only for planning and a couple of turns.

• The one you should always use besides Sol-Ultra is Luna-High.

• I rarely used Terra-Ultra because it’s hard to know the “middle tasks”.

Hopefully, I cleared up a bit of the confusion around the new ChatGPT, the new models, and the new capabilities. It’s all fun and games, but IS IT WORTH IT???

3. “You’re out of Codex and Work usage”.

I pay for ChatGPT Business & Claude Business plans.

Well, during my test of ChatGPT Work… I had almost no work done.

Because I kept seeing this:

“You’re out of Codex and Work usage.”

I honestly don’t understand why.

I can’t get anything done…

It went like this:

• I planned the task once with Sol-Ultra.

• Then I used Luna-High to do the tasks (just like I use Claude Fable-5).

• And I’m out of usage… in the middle of the task (!!!).

Even writing this newsletter was a challenge.

I couldn’t get anything done without hitting my limits.

Then I thought: “OK, maybe it’s because I pay $20 for ChatGPT, but I pay $100 for Claude. Let’s try to upgrade my subscription.”

You literally CANNOT get a better ChatGPT subscription.

On the flip side, Claude lets you pay $100/month if you want to:

OK, but let’s say you want to switch to “ChatGPT Enterprise” to avoid hitting your limits. How bad is it? Let’s do a simulation with:

1. A company of 150 people (the minimum for an enterprise ChatGPT).

2. 10 new chats per workday per employee (so 10 x 5 days x 150 employees).

3. 10 turns per chat (you ask a question + ChatGPT answers = 2 turns).

4. Let’s say the first prompt is Sol (Ultra) and then only Luna (High).

5. The total cost of subscription is already $20 x 150 = $3,000 per month.

6. But the added cost of the new ChatGPT is… $17,000 per month.

I consult companies on their AI adoption.

I can’t imagine pitching “so it’s $20 per employee per month + $100-500 if they use it correctly and understand token management”. That would be crazy.

It’s much easier to say “it’s $100 per month per employee, virtually unlimited usage (besides Fable-5, which isn’t necessary for non-coding tasks).”

I offer discovery calls for enterprises of over 100 employees that need faster AI adoption (or any AI adoption at all). Send me a message for more info:

Now, to be extremely clear, Anthropic (Claude) also aggressively charges you beyond 150 seats. It goes like this:

• If you are at 150 Anthropic (Claude) seats, you can pay $20/$100 per seat and get virtually unlimited chats with Opus models. That’s what I do and recommend companies do.

• If you are at 151 Anthropic (Claude) seats, you pay per use (on top of a $20/seat subscription). The pricing becomes pretty much the same as ChatGPT. So $3,000/mo of subscription and easily tens of thousands of token costs on top of it…

It’s almost as if you should stay under the 150-seat mark. And I’m not the only one who can do basic maths:

Token-maxxing is coming to an end. You can’t keep up once you pay for every token.

Token-maxxing is how companies want their employees to use as much AI tokens as possible (so agressively using AI).

A quick recap:

• Claude up to 150 seats is a no-brainer because of the $100/mo tier for businesses. You have a lot of free usage inside the $100 tier.

• Claude & ChatGPT are neck-and-neck past 150 seats, and super expensive, too.

So can you still use ChatGPT for work?

Share

4. Can you still use ChatGPT for work?

I never really stopped my ChatGPT subscription.

First, because I can’t. It’s my freaking job aha. Over 800,000 people are reading my newsletters every week, so I better be on top of AI, always.

Second, because I love ChatGPT for its image generation. It is still the best on the market, by very far. Most of my newsletter covers are made first on Midjourney and then edited on ChatGPT. Maybe it’s time for me to short Adobe.

OK, now about the new models (especially Sol, the smartest one) and ChatGPT Work (the Cowork killer), this is what I won’t tell you about them:

1. They are smart and capable of doing similar things to Claude.

2. You can make PDFs, edit files on your computer, make Excel…

3. You can vibecode the same way Claude does.

4. The difference is very slim. Too slim to care.

You are not reading a newsletter to know what you can do with it; you want to know what’s different about the model & the new ChatGPT mode.

Let’s only talk about the differences:

#1: ChatGPT has templates.

When you click on “Plugins,” you can find templates:

So far, again, nothing too different.

I gave it a task, and it took a whopping 16 minutes to complete it on Standard (not the Fast mode, which is 1.5x more expensive). I edited the full video:

The result? I was happier with the Claude (Fable-5) version than the ChatGPT (Sol-5.6) version:

As always, the best test is YOU. Use a personal ChatGPT account, download the app, and check for your case study which one is best. I can’t answer it better than you. I mean, I can; this is what my consulting firm does for businesses.

You can also create your own templates inside ChatGPT. I guess that’s a bit different from Claude:

But then what’s the difference between a template and a skill? I don’t know; you tell me. Well, actually, I asked ChatGPT:

#2: ChatGPT has Skills.

Claude too. But let’s test the ones from ChatGPT:

Good news: you can transfer any Claude skills to ChatGPT.

Prompt: make a skill from this skill, just update "Claude" to "ChatGPT"

Nothing (very) different from Claude.

I tried to make it different, I promise.

#3: ChatGPT has Sites.

This time it’s really different!! It’s a new tab:

I instantly wanted to try it on important matters, like solving cancer:

It took something like 15 minutes from start to finish, but I’m happy with the result.

It was cool. A fun experience. You can imagine many more capabilities in the future, kind of like Lovable (and the other vibecoding tools).

But for now, ChatGPT Sites is not as good as ChatGPT Codex or Claude Code to make your entire website from start to finish.

As AI gets better, it will be closer to ChatGPT Sites than to coding (you don’t see any “coding” element anywhere but only the end result, right inside the internet).

5. So ChatGPT or Claude?

Claude.

As you have already understood, they are very similar.

But my company has <150 employees, so the $100/month tier is way, waaaay better than the $20/mo tier from ChatGPT. We pay for a single shared ChatGPT account ($20/month) for image generation.

What if you have over 150 people in your company?

Well, it sucks. I would look into ways to stay under the 150 mark. 151 employees (and over) make AI very expensive to run with closed labs (Anthropic/OpenAI).

OK, but ChatGPT or Claude?

Over 150 employees, it only depends on what your team is comfortable with.

Your entire team loves Claude? They should stay there.

Your entire team only understands ChatGPT? Go there.

Because the real opportunity cost is not Claude vs. ChatGPT.

It’s paying for AI your team doesn’t touch. Adoption is the whole game.

When my team consults businesses, they never lack knowledge on how to use a tool. There is always a Superman of AI internally, the AI champion. The problem is how do you get 150+ Supermans, and can you even get there?

My core belief is that some people love the technology. They love it so much that they will constantly push themselves to try, test, and get better at it (here, AI). They don’t count time or energy on it. Kind of like someone collecting stamps. They just love the game.

Your entire team won’t be AI champions.

Just like your entire team is not the best at Excel. But they should be able to 1/ open it 2/ understand it 3/ work with it 4/ edit it if needed.

Who pays you, Claude or ChatGPT?

None.

I want to help millions of people know how to use AI. I can’t be paid by the AI labs I talk about every single day. One because it would be low-trust for you. Second, because I would need approval for every word.

I write constantly. I test everything. I can’t be stopped by contracts.

I do have partnerships with Gamma, Granola & Wispr Flow: I love their slides, meeting notes and dictation tools respectively. You don’t have to use them though.

But which AI am I using every single day? That’s a bigger one. So I use Claude. For now. And I am ready to switch when the jump is so big & obvious I must.

This newsletter is free, and will remain free. It grows from your shares, which are also free. I am forever grateful for your 2-second support that keeps How to AI free:

Share

I do train businesses, teams of over 50 people, on whichever tool stack you guys are using. The goal is to get stronger & faster adoption without getting a million-dollar token bill.

Send me a message if you want a discovery call with my partners:

You have recommendation on my next newsletter?

Leave a comment, and my AI will compile all of them:

Leave a comment

Upgrade your subscription to join our Circle community:

Subscribe now

Or it will cost you an arm and a leg.

Smarter AI = Expensive AI.

And the new Claude Fable-5 is by far the smartest AI model on Earth.

I mean this smart:

OK. So what can you do with this ‘Claude’?

Pretty much anything:

“OK, but this is not new. The old Claude can also do it.”

You’re right.

But you need to understand the difference between the many Claude.

Claude comes in different flavors of intelligence:

1. Haiku, the fastest but dumbest model.

2. Sonnet, the middle one. Good fast/smart ratio.

3. Opus, the big one. Very smart, used to be the smartest.

4. Mythos, the new smartest model. But also the most expensive.

Then each flavor is associated with a number, kinda like iPhone 6 vs iPhone 17.

But somehow, just to confuse you a bit, they launch a “Mythos-level” of Claude with a name associated with it instead of just a number.

And it’s called “Fable-5”.

The only thing you need to remember is that Fable-5 is the smartest AI around. It’s actually so powerful; you must experience it sooner rather than later.

“So, OK, what’s a “smarter” AI in practice?”

Think of Claude as an expensive senior lawyer and their intern.

• Fable-5 is the expensive senior lawyer with 20 years of experience. He costs $1,000 per hour.

• Sonnet-5 is the cheap intern, and only costs $100/hour. Worth noticing, he can still read what Fable-5 tells him to do.

You don’t want to pay the expensive lawyer for every quick contract review. You want to pay the expensive senior lawyer to define the overall international contract strategy. Then the drafting assistant (Sonnet-5) follows that strategy consistently.

Now imagine you lead a team of 100 people.

They all have Fable-5 and zero training. So they spend dozens of hours billing the senior lawyer for the most mundane tasks, and no one tracks it.

Well, you will soon wake up with a $,000,000 bill from Anthropic.

Not fun.

You - and your team - must know:

• When to use Fable-5, and when NOT to use it.

• How to best use Fable-5 to get the most of it.

I won’t waste your time explaining why this model is widely better than the previous one. You trust me.

I will also assume you never tried the new Claude 5.

And by the end of this (very long) guide, you will know more than the entire world. Because yes, 99.64% of the world never tried Claude.

Sounds like a good deal? Good.

But two things before we start:

1. Save this guide. Block 10 min this week & try Claude-5 for the first time.

2. Send it to anyone who has never tried Claude (& still thinks it’s a French name).

Share

PS: Think of one person still typing everything into Google like it’s 2021. Send them this. You’ll be the friend who finally got them using AI right. People remember.

1. Don’t prompt Fable 5.

Fable-5 now costs extra, starting July 12th.

Here’s the timeline:

• June 9: Fable-5 launches, included in paid plans.

• June 12: banned by US government. Gone worldwide, overnight.

• July 1: it’s back, included again (up to 50% of your weekly limits).

• July 12: it leaves subscriptions. From now on, Fable runs on usage credits. You pay per use.

The price: $10 per million input tokens, $50 per million output tokens.

In normal-people terms…

• $0.15 for a one short question, and one answer (of 1,000 words).

• $6 for 19 turns of long questions & answers (of 1,000 words).

• $14 when you hit 40 turns. Turns are the most expensive habit.

Here’s how to understand it:

• You have turns (your prompt + Claude’s answer = 2 turns).

• Each prompt (you) or answer (Claude) is using tokens for words.

• The more tokens, the more expensive it is.

• So the longer the prompts/answers or the more turns = the more $$$.

• And also, Claude reads the entire thread at every turn. So it scales fast.

Here’s the math:

The longer your chat is, the more expensive it becomes.

So what can you do?

1. Ask Fable super hard goals, for one or two turns max.

2. Switch to Opus 4.8 (High) for the rest. Because it’s not pay-per-use.

It’s super simple. Here’s my explanation with screenshots:

I need [task] for [goal]. I will expect [goal] to be achieved once we hit [specific targets]. Start by asking me questions about [task, goal and targets] to fully understand the context, using the AskUserQuestion tool.

That’s the right way of using Claude Fable-5.

You use it, few turns, and then you switch the model to Opus.

How do I know we must switch models?

Because I read their entire documentation:

And I am also chronically online on Twitter. I have your back.

You still need to know two things before I can move on to more complex concepts.

1. What happens if you only use Fable-5?

2. After using Fable-5 once or twice, should you switch to Sonnet or Opus?

Let’s start with the first one.

#1. I did the test.

And I had to spend $14 (!!!), one-shot, on top of my $100 plan.

Now I do have to say, Fable-5 wrote 30 entire newsletters, right inside my computer:

If you need to check your spending and limit your team’s credits, you need to go in your organization settings, in Usage:

This section is the most important one, right below:

Anthropic did say Fable returns to subscriptions “as soon as capacity allows.”

So am I telling you to close this guide because AI is too expensive?

The opposite. Here’s how you play it:

1. Sonnet and Opus remains included in your subscription. It handles your daily work, and it’s better than anything you had 6 months ago. Everything in this guide works on Sonnet or Opus, after using Fable.

2. Aim Fable at expensive problems only. Paying per token to “rewrite this email” is shooting a bird with a bazooka (please leave birds alone, though).

I wrote a newsletter about saving tokens if you have another 10 minutes to spare. But long story short:

• Create a new chat for a new task.

• Don’t talk too much to Fable-5 in one long session.

• Switch from Fable-5 to Sonnet/Opus during the session.

#2. Sonnet or Opus?

I have no idea.

I’m guessing from my own experience.

• On the one hand: Anthropic wants you to use Sonnet 5. Like they said it eveeeerywheere. But it’s their latest model (besides Fable).

• On the other hand: every benchmark shows Opus 4.8 as cheaper & better than Sonnet 5. Soooooo… I prefer cheaper and smarter, no?

So far I’ve been using Fable-5 (High) + Opus 4.8 (High).

Stay subscriber to my newsletter, and by next month, I will share my Anthropic bill.

Subscribe now

What should be our max per month? $200? $1,000?

Superintelligence is getting expensive…

2. Expensive superintelligence.

You need to pay for Claude (at least $20 per month; I personally pay $100) to access Fable-5. The free Claude won’t get you anywhere anyway.

Then you can pick the model’s “Effort”:

The effort level is just how long it takes before answering you.

The longer it takes, the more credits it takes.

So you might be wondering, “Why are we even using Fable-5 then?”

Well, it’s better at literally everything.

Here’s an example. Copy-paste this exact prompt:

Ask me what I do for work and what I keep track of in my head all day. Then build me a small interactive tool I'd actually use every week. You pick what to build based on my answers. Make it look expensive, and add one feature I wouldn't have thought to ask for.

Claude will ask you questions. Answer them.

Then watch the right side of your screen.

That window is called an Artifact.

Anything Claude makes that’s bigger than a chat reply opens there: documents, websites, mini-apps, dashboards, charts, even small games. You can open it full screen.

You can publish it as a link and send it to your team:

Sometimes an artifact can be interactive too, inside the chat.

Remember the orange image at the top of this newsletter?

Made with Claude.

But you now know you need to balance out Fable-5 (planning, or a quick super-good results within 2 turns) with Sonnet-5 or Opus 4.8.

OK, good. But how do we prompt this expensive superintelligence?

It’s been 4 years we have AI, and people (you) got away with these prompts:

• “Rewrite this email.”

• “Summarize this PDF.”

• “Make it shorter.”

Tasks. Small ones.

One step, then you grab the wheel back, then another small step.

These prompts made sense, especially between 2022-2024.

Old models lost the plot after 3 steps, so you learned to chop your work into bite-sized prompts. I probably wrote 100+ of these guides.

But Claude Fable-5 is different. Another league of its own.

It plans. It asks you questions before starting. It works for a long time without drifting. It pushes back. It checks its own work before handing it back, like an employee who re-reads before hitting send.

So here’s the new rule:

Give it goals. Long ones. Hard ones. Stop giving it tasks.

✘ A task: “Write a follow-up email to this client.”

☑ A goal: “This client owes me 2 invoices and went quiet 3 weeks ago. Here’s the full thread. Goal: get paid without burning the relationship. Plan your approach, draft whatever’s needed, and ask me what you don’t know.”

How do I even know this?

Because I read the docs for you.

An Anthropic’s engineer published another guide on how to talk to Fable-5.

It’s written for developers. Nobody told normal people what to do.

Here are the most important points inside:

Line 1. kill the buried answer.

Lead with the outcome. Your first sentence should answer “what happened” or “what did you find.” Details after.

No more scrolling through methodology to find the actual answer. Every response starts with the bottom line.

Line 2. the anti-bullshit line.

Only report work you can point to evidence for. If something is not yet verified, say so explicitly.

This one is for long tasks. In Anthropic’s own testing, this line nearly eliminated inflated “all done!” reports, even on tasks designed to provoke them. Same standard you’d hold a new hire to: don’t tell me it’s done, show me.

Line 3. draw the line between advising and acting.

Fable is eager. It will write the email when you just wanted feedback on it. So say which mode you’re in:

I’m thinking out loud. Give me your thoughts, don’t write anything yet.

or

Handle this end-to-end. Only check in with me if something is irreversible or genuinely needs my input.

Line 4. the anti-overthinking line.

When you have enough information to act, act. Give me a recommendation, not a survey of options.

For when you want an answer, not a menu.

You have the rules. Now time to create some stuff.

Share

3. Some Fable-5 worthy tasks.

If you handed me an AGI (an AI smarter than any human), I wouldn’t know what to ask for. What’s my AGI-level question?

Some obvious ones:

• Cure all diseases.

• Make me very rich.

• Fix everything wrong in the world.

If you’ve watched the super-duper viral movie “Obsession”, it feels like the make-a-wish object that quickly fires back…

So before we start curing cancers with AI - and some researchers are already on it, creating new drugs at record speed - let’s try to give Fable-5 some worthy prompts to use it best.

1 - Help me decide prompt.

Why it’s Fable-worthy: it forces the model to hold your stated priorities, run bear/base/bull scenarios, and argue against its own recommendation. I like the multi-angle reasoning. Usually, older models flattened into “it depends.”

I don’t want “it depends”, I need to take a sound decision.

Copy and paste this:

I’m about to make a significant decision and I want it stress-tested before I commit, because reversing it later will be expensive. Treat this like the decision review a sharp board member would run — not a summary of considerations.

First, ask me for: the decision and the options on the table, my priorities ranked in order, the constraints I can’t change, and the deadline. Then interview me — a few questions at a time — until you understand the decision better than I’ve articulated it. Push on anything I’m being vague about; vagueness is usually where the risk hides. Once you have enough, deliver your verdict. Open with your bottom-line recommendation in one sentence.

Then: how each option scores against my ranked priorities, the best case, base case, and worst case for your recommended option, the strongest argument for the option you rejected, and the single most likely way your recommendation turns out to be wrong. Be direct. Don’t hedge with “it depends on what you value” — I will have told you what I value. If my priorities contradict each other, name the contradiction instead of working around it. Before you finish, check your reasoning against my stated priorities and fix anything that drifted.

Use the AskUserQuestion tool for any question.

Here’s an example from this morning:

2. The Deep Research prompt.

Why it’s Fable-worthy: end-to-end research plus self-verification is exactly the long-horizon, evidence-grounded work Fable 5 was built for.

Here’s the prompt:

I need a research report I can actually make a decision from, not a survey of everything that exists. It will be read by someone busy who will act on it, so wrong claims are worse than missing claims.

Start by asking me for: the question I’m trying to answer, the decision it feeds into, and any sources I already have or trust. Then research it end to end. Use the AskUserQuestion tool for it.

Done means a report that opens with the answer in three sentences, then supports it. For every substantive claim, cite the source. Where sources disagree, say so and tell me which you’d trust and why, rather than averaging them into mush. Separate what the evidence shows from what you’re inferring — label the inferences.

Before delivering, run an adversarial pass on your own draft: attack your three most load-bearing claims as if you were a skeptic paid to find the flaw, and revise anything that doesn’t survive. If a claim can’t be verified, keep it — but flag it plainly as unverified instead of quietly dropping the caveat.

Keep it under two pages. Everything cut should be detail that wouldn’t change what the reader does next.

And my own example:

3. The Archive-to-Argument prompt.

Why it’s Fable-worthy: it explores a large, messy corpus, proposes competing narrative shapes, and pauses at exactly the point where your judgment matters.

Because I don’t trust complete-autopilot-AI yet.

Yes, even me. The prompt right here:

I want to develop a piece of writing from raw material — notes, transcripts, drafts, research — where the argument hasn’t taken shape yet. Your job is to help me find the piece inside the material, not to write a generic article about the topic.

Ask me to paste or attach the raw material, plus: who the piece is for, where it will be published, and what I want a reader to understand or feel by the end.

Then explore before drafting. Identify the most interesting tensions, surprises, and unresolved questions in the material. Ask me about the judgment calls only I can make — what I actually believe, what I witnessed, what I’m willing to say publicly. Propose three genuinely different arguments the material could support, with one line each on what that version emphasizes and what it sacrifices. Then stop and wait for me to choose.

After I choose, write a full draft in the voice of the raw material — my phrasing, my rhythm — not a polished house style. Flag every place where the draft asserts something the material doesn’t support, so I can either supply the evidence or cut the claim. Don’t invent quotes, anecdotes, or numbers to fill gaps; leave a visible gap instead.

And it works:

4. The Operating Review prompt

Why it’s Fable-worthy: multi-source synthesis across formats — notes, threads, action items — where the model has to reconcile conflicts and route judgment calls back to you.

Again, I love using smart AI for super messy context.

The prompt:

I’m going to give you the raw exhaust of my work week — meeting notes, message threads, to-do lists, half-written docs — and I need it turned into an operating plan I can run next week from. The point is to stop losing commitments and decisions in the pile.

Ask me to paste or attach everything, plus my top priority for the next two weeks.

Done means one document with four sections. Commitments: everything I promised someone or someone promised me, with who, what, and when — flag anything already overdue. Decisions: what was actually decided this week versus what was merely discussed; if the record is ambiguous about which, put it in the ambiguous pile rather than guessing. Conflicts: places where two meetings, people, or documents point in opposite directions — quote both sides. Next week: the five actions that most advance the priority I gave you, each traceable to something in the material.

Ground everything in the source — if I ask where an item came from, you should be able to point to the line. Don’t pad the plan with sensible-sounding actions that appear nowhere in my material. If the week’s material contradicts my stated priority, say so at the top; that’s the most useful thing you could tell me.

Pro tip: Connect your Granola to get your meeting notes inside Claude.

Ah sorry - you don’t know how? Here’s how:

4. Connect Fable to everything.

I pay for Claude with my company.

So I can safely connect all my apps without worrying about sharing data with Anthropic.

This is how to connect your Claude to the rest:

I showed you the example with Gmail.

But connectors are the best way to power up Claude.

Here’s my favorite list:

Connecting Claude to apps gives it context.

Because Claude doesn’t know you.

It read most of the internet. It never read your week. Your clients, your deadlines, your writing voice, the deck you half-finished on Sunday: invisible.

Every new chat starts at zero.

Context is everything you hand Claude before it works. Your prompt. Your files. Your examples. The earlier messages in the chat. The tools you’ve connected (Gmail, Drive, your folders).

And context decides everything, for one simple reason:

AI is average by default. It was trained on everyone, so it answers like everyone. Generic in, generic out. Context is how you drag it from “everyone” to “you.”

Don’t prompt…

“Write me a LinkedIn post about productivity”

But prompt…

Paste 3 of your old posts + “here’s my draft idea, write it like me”.

Now the pro move is to stop re-typing your context every time.

You have options, from easy to powerful:

Option 1 - Paste it. Keep an “about-me” google doc. Paste it at the start of any chat that matters, like a pre-prompt.

Option 2 - Projects. A folder inside Claude that remembers your files and instructions across chats. Super easy to use:

• Your chats (bottom left) = your chat history. And Claude remembers that.

• Instructions (top right) = a prompt that Claude reads everytime.

• Files (middle right) = the most important part. Upload PDFs, excels, docs…

Option 3 - A real folder. On your computer, for Cowork.

Share

5. What the f*** is Claude Cowork?

I am sorry to tell you it was last week’s newsletter.

So go and read it:

I will wait for you patiently.

And once you know Cowork, you will want to know Claude Code:

6. What the f*** is Claude Code?

Claude Code is thing developers won’t shut up about. The #1 reason people pay (a lot) for AI is to code.

But for rookies like you & me, we care about vibecoding.

And I wrote a free guide on it:

Alright, sorry for these articles to read.

Back to Fable-5 guide. I made a skill you can download to prompt it better:

Share

7. /fable-prompter

I spent 4 years learning how to prompt.

You get to skip that (but you still need to understand what’s going on).

First, what a skill is: instructions you save once, and Claude loads them whenever you type their name. You type /name + one line. Claude does the rest. Skills work in the regular chat and in Cowork (and in Code).

Skills exist because repeating yourself is a waste of context (and of your life). Anything you’ve explained to Claude twice should become a skill.

So I built one called /fable-prompter.

You hand it a lazy, half-formed idea. It hands you back a finished, ready-to-send prompt, built exactly the way Fable-5 likes them. Specific. Structured. Reasoning dialed all the way up. You copy it into a fresh chat, hit send, done.

The most important part is to give it to Opus 4.8 (not Fable).

I typed 9 words:

It returned a 15-line prompt: audience defined, hook rules, voice rules, length caps. And it ends by telling Claude to ask ME for the missing details before drafting. I wrote 9 words.

So now you can copy this and paste it into a Fable-5 instance.

How to install a skill (the easiest steps on Earth):

1. Download the /fable-prompter file.

2. It’s on Dropbox, you will need a password.

3. Put: RUBEN-HOWTOAI

Leave a comment

4. Go to claude.ai → Settings → find Skills (under Capabilities) → click + and upload the file.

5. Toggle it on. Open any chat. Type /fable-prompter + your idea.

By the way, I gave a library of my favorite skills to Claude.

It’s at my welcome email for people who are subscribers (it’s free).

It looks like this:

If you are already a subscriber, leave a comment & I’ll dm you:

Leave a comment

8. Your team’s brain, in a folder.

Your best teammate quits tomorrow. What leaves with them?

Everything.

The way they brief clients. The pricing logic. The 14 lessons they learned the hard way. It all lives in their head and their DMs.

Companies answer this with SOPs. Documents nobody reads, written once, outdated by next quarter.

I could teach you how to save someone’s brain. But 1/ it takes too long, 2/ it’s too industry-specific, 3/ that’s why I built a company around it.

We go into companies, help them extract knowledge from key people.

Now hold up. I see the haters coming at me. “Ruben, you help companies fire people ????” Hum. No. We helped people retire. They stopped working because they wanted to retire but couldn’t because they were afraid of sinking the company.

We call it the Wisdom Walkout.

Here are two quick stories that may sound familiar:

1. CEO of a family metal processing business knows every customer and metal without a spreadsheet in sight. We interviewed him and reviewed 2 decades of emails. Now anyone can chat with Jim’s AI Twin.

2. Global engineering design and manufacturing firm has 60 out of 350 employees over retirement age. They have a ticking time bomb of relationship data that’s about to walk out. We are currently indexing their team communication, paired with human-to-human interviews, to create a central knowledge base.

You feel seen? Send me a message to book a discovery call:

I don’t know nothing.

I’m not technical.

I dropped out of university.

I ran a techno music label before all of this (from being 17 to being 24).

If I can hand my week to an AI and get it back with the work done, so can you.

Be part of the 0.36% who actually tried Claude.

Heck, be part of the 0.36% who is doing it well!

I wish to help a tiny percent of humanity to know how to get better at AI. Even 1% of us. That’s like? 82 million people? I’d love to. Help me get there, share it.

Share

PS: This newsletter stays free because you guys share it. Every article crosses 1,000+ shares, and it’s my weekly north star. The best share is your team’s group chat (Slack or Teams): you save them tokens, and you help me spread the word.

I send a welcome email gift to new subscribers. But I created it when I already had 500,000 subscribers. So if you are one of them and you never received the gift, leave a comment and I will dm you.

Leave a comment

I wrote “Cowork” and “Cowork 2.0.” My most-shared work ever, across Substack, LinkedIn, and X. A total of 20 million views.

Do not read them.

To be totally fair, I was right about one thing: Claude will become bigger than ChatGPT because of the launch of Cowork in January.

Cowork is the best thing to happen to AI since ChatGPT.

If you don’t code, you must be using Claude Cowork.

Since then, I have written hundreds of guides and two major articles. I shared how to set up Cowork, and that was the latest (outdated) version of it:

The setup is good…

Until you go to enterprises, and they need to collaborate.

I started giving workshops to enterprises. Teams with hundreds of people using Claude Cowork together. And my files + folder system broke.

How Anthropic wants us to use their AI is different from how people actually use it.

And that’s OK.

Because I’m not paid by Anthropic.

So I can say whatever I want on how to best set up Cowork for you + your team.

I spent countless hours updating my setup. This guide shows you exactly what & how. By the end of it, you’ll know how to start - and what to absolutely avoid - with Claude Cowork.

Before starting, I want you to do two things:

1. Save this guide & block 20 minutes this week to set up Cowork.

2. Send it to anyone who still hasn’t tried Cowork (or Claude).

Share

Skip this if you already use Cowork.

Quick reminder on how to access Claude Cowork:

1. Go to claude.com/download. Download the app on your computer.

2. You must have a Pro account ($20/month). I pay for the $100/month plan.

3. Open the app. Click on the Cowork tab at the top between Chat & Code.

4. You can add Skills and start Projects on the left menu. More about it later.

5. Make sure to select “Opus” for complex tasks. It’s the default model.

6. Once Fable is available, use it. It’s (somehow) even better.

I. Cowork is many Claude.

Cowork is much better than the normal Claude.

In simple terms, why is Cowork better?

Because it answers with many Claude at once.