How to work while you sleep (Agentic AI)

May 07, 2026

Before we get to today’s post, I have exciting news. I’m launching the Ultra Labs 1:1 executive coaching program. In it, I help executives, entrepreneurs, and leaders unlock next-level productivity and scale themselves with agentic AI in just 3 months.

You’ll get tailored systems, support, and accountability to perform at your best without burnout, turn agentic AI into real leverage, and put it all into practice. It includes 1:1 coaching with me, 10/6 access to my phone (10 hours/day, 6 days/week), and a guaranteed 100% productivity increase.

The Ultra Labs 1:1 executive coaching program is our most exclusive offer and is limited to 10 seats at a time. If you’re interested in one of the next available seats, learn more and book a free call here.

"Earn with your mind, not your time" — Naval Ravikant

Why it works

Naval Ravikant (pictured) is one of the most influential founders, investors, and thought leaders of our time. He is the co-founder of AngelList and was an early investor in companies like Uber and Twitter. His ideas on creating wealth have reached millions and were distilled into a book by Eric Jorgenson entitled The Almanack of Naval Ravikant, which is one of the best books I know on succeeding in the 21st century. You can read it for free here.

Among Naval's most influential ideas are the four forms of leverage: people, capital, media, and code. The first two are older forms of leverage that humans have used for millennia. People means having others work on your behalf. Capital means putting money to work in your place. Both are what Naval calls permission-based. To employ people, you first need someone willing to follow you. To deploy capital, you first need someone willing to provide it to you. The two new forms of leverage are media and code. Media is content, like a social media post, podcast, or a newsletter. Code is software that works for you while you sleep. Unlike people and capital, media and code are permissionless. You don’t need anyone's approval to publish a social media post, send out a newsletter, or build software.

Today's newsletter is all about code. For decades, it has been the most powerful form of leverage, but it used to be gated by one thing: you needed to know how to write it. With the advent of agentic AI, this form of leverage is now available to anyone who can describe what they want in plain English (or any other language). As I am writing this, Anthropic's Claude is arguably the most powerful, usable, and secure tool available for putting agentic AI to work. Today's post is about making Claude work for you in three simple steps: setting it up, testing it, and using it.

How to do it

1) Set up Claude
Out of the box, Claude lacks three things: memory, maximum intelligence, and personalization. A one-time setup turns it into a tool that knows what to draw on, how deeply to think, and how to work with you. Here are three ways to dial it in before you do anything else.

Tip 1: Activate memory (and privacy)
Start by turning on memory. Under Settings/Capabilities, switch on both "Search and reference chats" and "Generate memory from chat history." The former lets Claude pull from conversations you have already had. The latter helps it remember useful context from previous chats. Together, they stop you from re-explaining who you are every time. Also, turn on privacy. Under Settings/Privacy, switch off the "Help improve Claude" toggle. This makes sure that everything you share with Claude stays between the two of you, instead of being used to train future models. This is activated by default for new accounts, so you now have to deliberately switch it off.

Tip 2: Upgrade intelligence
Click the model dropdown in the input area and pick the latest model available. Each new generation tends to be meaningfully smarter than the last, and on harder, multi-step work the gap can be several times larger, not a few percent. In the same dropdown, make sure Adaptive Thinking is toggled on. When activated, Claude decides when to think harder. Quick questions get fast answers. Complex tasks get deep reasoning. The combination of the best model and adaptive thinking is the difference between a quick assistant and a real thinking partner.

Tip 3: Add personalization
Tell Claude about you. Open a new chat and say the following: "I want to write my Instructions for Claude. Please interview me by asking 5 questions about me, my work, and what I want from you." Answer the questions in detail. This will take a couple of minutes, but I promise it is worth it. When you’re ready, say the following: "Now turn what you learned from my answers into a text that I can paste into the custom instructions block." Lastly, copy what it gives you, go to Settings/Profile, and paste the text under "Instructions for Claude." From now on, every conversation starts with Claude knowing the essentials about you.

2) Test Claude
After Claude is set up, do not just assume it works. Test it. The mistake most people make is to throw a high-stakes task at Claude, watch the first draft fall flat, and conclude that it does not work for them. Don’t be most people. Here are three quick tips to close the loop fast.

Tip 1: Test whether it knows you
Open a fresh chat (so your instructions load cleanly) and ask: "What do you know about me and how I work?" If Claude leads with your role, your work, and specific preferences you shared with it, your instructions are landing. If the answer is generic, your instructions are too thin. Add more detail about what you do, what you are working on, and what you want Claude to remember.

Tip 2: Test whether it sounds like you
In the same chat, ask: “Draft a short email in my voice to my colleague X and ask about Y.” If it reads like something you would actually send in your tone, length, and phrasing, you are in good shape. If it sounds cold or generic, add concrete rules, such as “keep it warm and brief” or “avoid corporate filler.”

Tip 3: Update your instructions
After going through tips 1 and 2, ask Claude: “Based on what I just told you, what should I add or change in my Instructions for Claude to fix this?” Claude will draft the update. Copy it and add it to the end of the text at Settings/Profile/Instructions for Claude. Open a fresh chat to confirm the new version loaded, and re-run the tests. Treat your instructions like a living document and improve them over time.

3) Use Claude
Once you trust the setup, the third step is to put Claude to work the right way. Treat Claude as a smart, fast junior, not a senior colleague who will catch your mistakes. Why? Because it usually won’t. It likes to please. Here are three tips for pulling that off.

Tip 1: Clarify and outline first
Always end your first request to Claude with “Ask me any questions before you start if anything is unclear.” This one line cuts assumptions and bad first drafts. The time you spend letting Claude interview you and answering those questions will save you far more time reviewing and rewriting later. For anything substantial, ask for the outline before the full output. Edit the outline, then have Claude build from there.

Tip 2: Never accept the first draft
Remember that AI likes to please. All models are designed to be helpful, which means you still need to direct, challenge, and review the output. So once Claude delivers, push back. Tell it specifically what is wrong, like "too formal" or "too long.” Generic feedback gets generic fixes. And when you land on the final version, paste it back into the chat and ask Claude to remember what good looks like to you.

Tip 3: Converse, don't "prompt"
Forget fancy prompt-engineering frameworks. AI is great at inferring what you already understand, sensing the level behind your question, and meeting you exactly where you are. Treat Claude like you would a smart, fast junior on your team, and explain things clearly and concretely. State who the work is for, what you need, the format you want, and what to avoid. Plain English, not a structured prompt template.

AI is all the rage these days, but here’s a harsh truth: 99% of people still use it as a chatbot. They ask it to write better emails, content, or presentations, and then they do the work themselves.

The three steps above already lift you above that. They turn Claude from a generic chatbot into a real thinking partner. But that’s still not where the leverage compounds. 

The real shift happens when you stop talking to AI and start delegating to it. Not a smarter assistant, but a cognitive workforce that takes multi-step work off your plate.

This is where Claude Cowork comes in. It launched in research preview in January 2026, became generally available in April 2026, and lets you put agentic AI to work in plain English, with no coding skills required.

Over the last few months, I've used Claude Cowork to help myself and my coaching clients save hours of tedious work each week and finally get to those tasks they or their teams never had the time for.

If you want to learn more about how I could help you do the same, book a free discovery call here.

Here’s my question for you this month:
What could you achieve if you had AI agents working for you 24/7?

And here’s my challenge for you:
Get started with Claude today. Download the desktop app. Set it up, test it, and use it. Then move to Cowork to go fully agentic.

It's still early days, and turning pro at working with agentic AI is one of the most important skills of our time.

Until next month,
Christian

PS:
Are you an executive, entrepreneur, or leader struggling with focus, energy, or stress? Are you curious about agentic AI but not sure how to make it work for you? Are you constantly putting off what matters most in your work and life?

If that sounds familiar, the Ultra Labs 1:1 executive coaching program is for you. In it, I personally provide the systems, support, and accountability to perform at your best without burnout, turn agentic AI into real leverage, and put it all into practice in just 3 months.

Since 2023, I’ve helped 120+ executives, entrepreneurs, and leaders unlock their full potential. The program includes:

  • 1:1 executive coaching with me

  • 10/6 access to my phone (10 hours/day, 6 days/week)

  • Tailored systems to unlock next-level productivity and AI-powered leverage

  • A premium device (OURA or WHOOP) to track energy, sleep, and stress

  • A guaranteed 100% productivity increase

Interested? Learn more and book your free discovery call here while you can. I only take on 10 clients at a time and seats are booking fast.

There are 800 million users on LinkedIn, and 99% of them aren’t creators.

Translation? There has never been a better time to grow an audience on LinkedIn.

Unlike most other LinkedIn growth coaches, my co-author Eric Partaker and I walked the talk:

  • We’ve built a 1+ million audience across LinkedIn (1,000,000+ followers) and our newsletters (200,000+ subscribers) in just 12 months

  • We’ve created a 7-figure business from our LinkedIn thought leadership through services (e.g. coaching) and products (e.g. courses)

  • We’ve been ranked #1 LinkedIn creators by Favikon and got a book deal with Portfolio Penguin for our first book

Our #1 goal for this coaching program is to make it the highest ROI investment of your life. Learn more and book a free discovery call here so that we can get in touch.

2 more ways I can help you

Want a free PDF of my top infographics? Get it here.

Interested in sponsorships for my newsletter or LinkedIn? Get in touch here.