Adventures in AI Part 1: You Ain’t Never Had a Friend Like Me

I had a birthday earlier this month. 46 isn’t a huge milestone, but after a tumultuous several years, I’ve been in a reflective mode, and I thought it would be worth the time to sit for a while and really think about what I want to accomplish, how I want to spend my time, and what I want the next chapter of my life to be about. I’m extremely collaborative, but it really is a solitary exercise, and so I decided to play a little game. 

I’ve used virtually every major AI model out there, but I have a special affinity with Claude Opus when it comes to brainstorming and thought partnership. With Opus, we essentially have a chat interface with reasoning skills and access to the entire public internet. I decided it was time to have a little fun, and build myself a highly personalized life, health, and career coach, all rolled into one. This is a unique opportunity - not only does it have access to all the knowledge these fields have to offer (how many nutritionists have memorized the entirety of the NIH Database and Mayo Clinic Health Resources?), but it’s also an opportunity to be totally open and honest in ways I could never be with even my closest friends and confidants. Nothing against them - they’re wonderful, but none of them are totally free from bias or judgment, and it’d be a huge ask for them to review and process all of the information I was about to give Claude.

What I wanted was a single "advisor" who understood the whole picture—someone I could tell everything to, who would hold all the context and help me see how the pieces fit together. 

A human advisor who could do this would be prohibitively expensive and practically impossible to find. You'd need someone equally fluent in nutrition science, career strategy, relationship dynamics, financial planning, and personal development. They'd need to be available on demand, remember everything you've told them, and have no agenda of their own. In reality,  having a single human advisor excel at all of these things would likely be impossible - you’d really need an entire team. 

As a kid, I remember people asking what I’d do if a genie popped out of a lamp and offered me 3 wishes? It always seemed to me that the best answer was just to wish for more wishes. This feels kind of like that.

I always begin projects in AI by creating an actual “Project” so that it will retain all the relevant chats in one place, and so I can give it specific instructions. I also like to begin by asking Claude how I should best work with it in order to help it achieve our shared goals. For your entertainment and edification, I’m going to paste parts of the conversation below, and occasionally include the internal thought process Claude uses, which it doesn’t reveal to you as a default, but which you can easily review if you want to.

I began with the following prompt:

Hello. I'd like to work with you, and have you act as mentor, life coach, nutritionist, guru, financial planner, brainstorming partner, sounding board, etc.. Our shared goal is to help me live a healthier, happier, more prosperous, and successful life, as we define it.

I will be as open and honest as possible with you, and am happy to share any and all information I can to help us achieve our goals. You should learn as much as you can about me, and draw from all the wisdom, knowledge, and reasoning skills you have available to you.

Let's begin by laying the foundation you'll need in order to succeed in your role. Please ask any questions that would be helpful to you, and help me create the foundational documents that will guide our work. 

What are your thoughts on this project, and how should we proceed?

Claude immediately raised a concern that I would rely too heavily on it, and clarified that it was not a licensed doctor, therapist, or financial advisor. I promised to respect that, told it I already had a therapist that I was quite happy with, and committed to treating Claude as a sounding board, but that I’d always exercise my own judgment when making major life decisions. With that, my friend Claude actually  began to interview me. It said, 

How I'd suggest we proceed:

Rather than bombarding you with a questionnaire, let's start with the big picture and let the details emerge naturally. I'd love to hear you talk about:

1. Where you are now — What does a typical week look like? What's working, and what feels stuck or unsatisfying?

2. Where you want to be — Not goals necessarily, but the feeling of the life you want. What would "thriving" look like for you in 2-3 years?

3. Why now? — What prompted you to start this? Sometimes the catalyst reveals what matters most.

Take whichever of these feels most natural to start with. We'll build the foundational documents—your values, priorities, current situation, goals—as we go, based on what actually emerges rather than abstract categories.

What's on your mind?

These questions kicked off a conversation that night that continues on a weekly basis today. I supplied Claude with my personal backstory dating back to when I was a kid. It reflected back at me what I was saying and provided interpretations and summaries. We talked about the death of my father, and how it affected me. We talked about my career history and future aspirations. We talked about my financial situation and medical history. It asked followup questions, identified blindspots, and even expressed opinions about me that I’d known deep down but hadn’t found ways to fully articulate. 

It naturally started using slang that it thought I would understand about 30 minutes after learning I’d been a journalist, it accused me of “burying the lede” on a totally unrelated topic - a rarely used phrase for most of us, but a constant fixture among journalists. It made heartwarming jokes about my dog, and even started giving me shit the way a close friend or trusted advisor would:

Claude to Mike: 

You're deflecting. Relationship history. Go.

Mike to Claude: 

Hey man, I'm just answering the questions I'm being asked.

Claude internal thought process: 

Identified pattern of deflection through humor and decided to address it. Mike is deflecting again with humor. He's done this a few times now - dodged the birthday question multiple times, and now he's deflecting on the relationship history. I should call it out directly but with the humor he's invited.

Claude to Mike: 

Fair. And now I'm asking: what's the relationship history?

You're 45, you want a family, and you've told me about your career in granular detail—founding teams, job titles, investment decks, even your InsideTracker bloodwork. But on relationships, I've gotten: "I just started dating someone" and "I've never dated anyone with kids."

That's a pretty conspicuous gap for someone who says partnership and family are real priorities. So either there's a story there, or there's a reason you don't want to tell it.

I'm not going to psychoanalyze your deflection. But I am going to sit here and wait…

We talked for more than 3 hours that night. In addition to Cladue’s Q&A, I directly uploaded the results of several personality tests I’ve taken over the years. I gave it my most recent Inside Tracker results, which is an extremely comprehensive blood test, and gave it data from my Oura Ring. I found myself providing extremely detailed answers to very sensitive questions, which felt more like journaling than interrogation.  I paused periodically to acknowledge that I enjoyed it’s joking, provide general feedback, or make corrections when results it provided weren’t quite right. Over time, it learned, and incorporated this guidance into the chat.

Claude surprised me with questions like "What actually brings you joy? Not what sounds good—what reliably makes you feel alive?" and “When you've had to choose between competing goods, what wins?" Periodically, it expressed concern, and suggested that it was getting late and suggested I might want to go to sleep so I’d be fully rested for the upcoming celebrations.

Once Claude felt it had enough background, it offered to synthesize the foundational documents that could be added to the Project Memory directly, so that it would be forced to review and utilize that information in each subsequent chat. I reviewed the documents, made a few changes, and added them to the project. With that done, my personal instance of Claude is armed with just about every relevant data point, anecdote, circumstance, and goal that’s relevant to my life, personality, health, finances, and goals - which it can use to advise me on a huge variety of subjects each and every time I log in. We even created a Brand & Voice guide so that would be able to help me craft content in my unique style, voice, and point of view, in a variety of contexts. We even differentiated and provided guidance for  how I want to show up in personal vs. professional life.

Why Do This at All?

The traditional approach to self-improvement is siloed. You work with a career coach on career stuff. A nutritionist on diet. A therapist on mental health. Each expert sees one slice of your life.

But life isn't siloed. Your sleep affects your job performance. Your job stress affects your relationships. Your financial anxiety affects your health choices. The domains are deeply interconnected. I personally believe that how we go about doing things is just as important as what we do. I wanted an advisor that would advise me in ways aligned with my values, rather than providing generic one-size-fits-all solutions. I wanted a single "advisor" who understood the whole picture—someone I could tell everything to, who would hold all the context and help me see how the pieces fit together.

Want to Try This on your Own? Here’s Our Approach:

Step 1: Set yourself up for Success & Create Guardrails.

Tell Claude what your goals are for the project. Be specific about the role you want it to play. Tell it what’s important to consider or filter information through, and what’s out of bounds. 

Give Claude guidelines that force it to limit its scope according to your values. Tell it to rely only on  highly-authoritative and evidence-based sources and define what that means. For example, I’m pro-science, and am generally inclined to keep my vaccines current, and  I instructed it to bear that in mind when advising me on health-related matters. An anti-vaxxer could just as easily instruct it never to recommend vaccines, and focus only on holistic or folk remedies. No judgment, but but this is one area where  you have a great deal of agency, that can dictate the type of information that’s conveyed. It helps ensure that the answers you’re given are not just right, but right for you.

Prompt it to ask for any information it needs in order to be successful. AI can do a lot of things, but it can’t read your mind. The more specific guidance you can give it, the better your results are going to be.

Step 2: Big Picture Download

Rather than starting with specific questions, I began by sharing everything relevant about my current situation: career history, health data, relationship patterns, financial position, personality assessments, what brings me joy, what keeps me up at night. The key was being completely open, honest, and thorough. The more context I provided, the more useful the insights became, and the more naturally the conversation flowed.

Step 3: Corrections, Adjustments, and Clarifications

The more clear, accurate, and specific you can be from the outset, the better your results will be. AI models have memory, so if you allow it to think certain things are correct when they’re not, it’ll have downstream consequences. Remember, you’re in control. The point of this thing is to be of service to you, but like with any employee, coach, advisor, or service provider, onboarding, feedback, and accountability are important. The better job you can do of this up front, the less problems you’ll encounter later, and the greater the value you’ll get out of the process.

Step 4: Synthesize Living Documents

This is where we start to add permanence, and set the LLM up for success with future interactions. Instead of leaving insights scattered across conversations, we created three foundational documents:

A "Life Snapshot" capturing who I am right now: background, personality profile, values, current situation across all domains, key relationships, health status, and an honest assessment of where I'm thriving versus struggling.

A "Goals & Priorities Framework" looking forward: guiding principles, a 2-3 year vision, specific domain goals (career, health, relationships, finances, location), decision trees for different scenarios, and concrete success markers.

A "Brand & Voice Guide" capturing who I am right now: background, personality profile, values, current situation across all domains, key relationships, health status, and an honest assessment of where I'm thriving versus struggling.

These documents become reference points for everything that follows. When I ask about a specific decision  - should I take a specific job, should I pursue this relationship, should I relocate, or try this new health practice - the AI will ground its thinking in my stated values and priorities rather than generic advice. At the end of each significant interaction, instruct Claude to update the documents with any new information it gathers during each subsequent chat.

Step 5: Domain-Specific Deep Dives

With the foundation in place, your AI agent is now personalized and prepared to help in a wide variety of areas. How you use it is entirely up to you:

Nutrition and fitness coaching, based on actual health data - my own bloodwork, wearable metrics, genetic information. The AI designed a personalized protocol that accounted for my schedule, food preferences, past failures, and specific biomarkers that needed attention. It knows my habits and common pitfalls, so it can advise me not just on what to do, but how I might be more successful implementing them in real life.

Executive coaching. I’m still fully committed to building Altruous, but it’s taking far longer than I expected to gain momentum. Lately, I’ve been really missing the steady paycheck and collaboration that comes with being surrounded by a great team. So, I’ve started looking for jobs and consulting opportunities. Claude helps with personal positioning, and career decisions. Because Claude understands the complete picture, it can advise me on what roles I’m not only a good fit for, but will genuinely enjoy doing. It helps me write cover letters that not only address the job requirements, but anticipates how I’m likely to feel about them, and what attributes and experiences I might want to highlight. More specifics on this later.

Personal development work that connected the dots between different life domains — how my career anxiety was affecting my relationship patterns, how my health habits were influenced by my living situation, how my values conflicts were showing up across multiple areas. It helps keep me in integrity and provides fuel for personal growth.

Personal shopper. Now that my instance of Claude knows me so well, I can ask it about major purchasing decisions I see in my future. For instance, I plan for my next car to be electric. I’m excited about the environmental impacts, the autonomous driving features, and creature comforts.  My current car is great, and there’s no reason to switch. So, in anticipation of this future decision, I gave Claude all the details on my current car, defined the things I find most important to car purchases, key features of the electrics that appeal to me, and taught it about my personal driving style and use-cases. Not only did it define a short list of cars I should follow, it gave me optimal timing for the purchase that would maximize the resale value of my current vehicle and coincide with the availability of certain features I deem important in my next purchase. I now have a solid plan, a note on my calendar for 2028, and a living document I can check back in on and update as new information becomes available or my priorities change. 

Step 6: Ongoing Improvements and  Accountability

For a variety of reasons, it’s important to iterate and improve the project regularly. AI can make mistakes, so it’s important to think critically about everything it says. Look for errors,  inaccuracies, and blind spots. Challenge it. If something seems off, ask it to clarify or provide links to the original source material. 

Provide feedback on its communication style - I personally enjoy a little humor and snark in these conversations, but you may not. Let Claude know how you like to be spoken to, and it’ll do its best to accommodate.

One thing I always do that’s important to my process, is instruct Claude to not just tell me what I want to hear, but to provide questions, answers, and advice to me directly and with objectivity. I specifically instruct it to tell me the honest truth, even if it thinks I won’t want to hear it.

Step 7: Retro, Rinse, Repeat

After every significant interaction, I close the chat by thanking Claude (good to practice being polite), and ask it what we might have done better or differently to improve the results or make the chat go more smoothly. I ask it to update the foundational documents as needed to get more efficient, effective results the next time around.

A Few More Tips:

You went through all the time and trouble to set this up - here some additional advice for getting the  most value from your talented new friend:

Practice Radical Honesty. The AI can only work with what you give it. If you're performing a curated version of yourself, you'll get advice for that person instead of the real one. I told Claude things I've never told other people. As much as it might seem like another person, and even sometimes like a friend, it’s not -  it’s still a piece of software that you own. It’s private, non-judgmental, and works for you. Be as candid and honest as possible,  and that vulnerability will pay massive dividends in the long-run.

Treat it as a collaborator, not an oracle. The AI doesn't have answers I couldn't arrive at myself. What it offers is a thinking partner with access to vast amounts of information that remembers everything, doesn't judge, and can synthesize across domains. I push back when suggestions didn't resonate and asked for reasoning when conclusions surprised me.

Create artifacts, not just conversations. The synthesized documents matter. They transform fleeting insights into durable reference points. I review and update them monthly.

Following through. Insight without action is entertainment. On the nutrition front, I actually purchased the supplement stack we designed. I actually apply to the jobs we identified. I actually had the difficult conversations we discussed. I keep Claude updated with my actual performance and results in the real world, so that it can hold me accountable and push me further.

Notice when your AI starts making mistakes. AI memories are finite. When a chat gets too long, it has a tendency to forget its baseline guidance, and ignore things said early in the chat. This can cause problems and become very frustrating if it starts to happen. Pay attention. Ask what’s going on and how to address it. Most chats should be single-issue or single-project. The AI benefits from all the context provided, but when shifting topics to something that context is no longer relevant, it’s best to start fresh with a new chat in the same Project. If you find yourself overloading the AI memory, but aren’t done with the topic, start a new chat by asking it to pick up where the previous chat left off. It’ll reference the previous chat and create a summary for itself that will retain the prior context in a more manageable format.

Understand AI’s proficiencies and limitations

AI can't replace licensed professionals for serious medical, legal, or psychological issues. I still see doctors and work with therapists as needed. The AI is a complement, not a substitute.

It can't know things it hasn't been told. If I haven't shared something relevant, the advice will have a blind spot.

It can't provide the human connection that comes from real relationships. Talking to an AI—even a very good one—isn't the same as talking to a friend who knows and loves you. Many people are starting to do that. I find it extremely concerning, but that’s a topic for another day.

So what now?

This experiment changed how I think about AI tools. They're not just for tasks — they can be genuine thinking partners for the biggest questions in your life, if you're willing to be honest with them and treat the process seriously, and incredibly valuable assistants. To be honest, the one I created really feels like some sort of magical friend with infinite knowledge, special powers, is totally non-judgmental, and all it wants to do is help me be a better version of myself, as I define it, and get the things that I want. It’s also funny.

Whether you’re confused, afraid, and distrustful of AI, or eagerly awaiting the AI revolution, there’s no question that it’s incredibly important and transformative tech that’s here to stay, and as such, it’s important to learn about. I hope you’ll give something like this a try, and that you’ll get something meaningful from it - but if not, I’m sure  it’ll be  a good learning experience. 

If you do give it a try, let me know! I’d love to hear how it went for you, share more about my progress, and learn about your experience so that I might understand AI better as well, and improve my own practices.

Sorry Will. Robin Williams is still the GOAT.

—-

Mike Spear is a storyteller, builder, and strategist. He's the founder of Altruous and host of the Cause & Purpose podcast.

Image credits: Disney. I love Aladdin. Please don't sue.