Last month, I went all-in on replacing my executive assistant with AI.
I bought a Mac Mini for $600. I burned through $700 in Claude tokens. I spent roughly 150 hours planning, architecting, and building. And I’m here to tell you — it’s far easier to utilize a great human executive assistant than it is to utilize an AI assistant.
Now, before you think I just don’t know what I’m doing, hear me out.
Why I Should Be the Guy Who Cracks This
I was a staff-level teaching engineer at Coinbase. They paid me hundreds of thousands of dollars a year to teach some of the world’s best engineers about blockchain technology. So I have the technical chops.
But more importantly, I’ve personally built the SOP and workflow catalog for over 60 CEOs through FreedUp. I’ve been on hundreds of CEO productivity coaching calls. I know how to design systems, document workflows, and delegate effectively.
If anyone should be able to make an AI assistant work, it’s me.
And yet - here I am, telling you it’s not what the internet wants you to believe.
AI Is a Powerful Tool. It Is Not a Helpful Assistant.
That’s the phrase I keep coming back to. In a limited scope, AI is incredibly powerful. It’s amazing at coding. It’s amazing at processing information. But when it comes to performing complex, multi-step workflows as your personal assistant? It’s incredibly difficult and finicky.
People see the power of AI in narrow use cases and extrapolate that into “AI will be your next executive assistant.” But there’s a massive gap between a powerful tool and a reliable helper.
Why Everyone’s Exaggerating (Including Me)
I think the AI assistant hype is inflated for two reasons.
First, people see what AI can do in limited contexts and assume it scales. It doesn’t - not yet. The jump from “summarize this document” to “manage my entire day” is enormous.
Second - and this is the one no one talks about - there’s a status game at play. Saying you have an AI assistant signals that you’re smart, technical, forward-thinking, and ahead of the curve. Nobody wants to be seen as redundant or deprecated. So we exaggerate what we’ve built and what it can do.
I caught myself doing this just Monday night. I’d gotten Open Claw set up on my new Mac Mini. I was at a bar with friends, super excited to show them I could text my AI assistant and it would text me back. Only problem? The Mac Mini went to sleep. It didn’t text me back. I was trying to flex and looked like an idiot because it was finicky and didn’t work.
Meanwhile, if I had texted my actual human assistant, she would have responded immediately. Nobody would have been impressed because it’s boring. But it works. Every time.
The Real Barrier Isn’t the Tech
Here’s my biggest takeaway from this whole experiment: the largest barrier to productivity with AI - or any assistant - is the same as it’s always been.
Can you design workflows? Can you systematize your life? Can you delegate?
If you don’t have a system for your email, you’ll always be reactive. If you don’t have a system for planning your week, you’ll always be reactive. If you don’t have a system for your calendar, you’ll crumble beneath it.
Being a CEO is incredibly complex. Too many variables. You need to simplify and assert control wherever you can, so that you’re free to be creative and flexible in the areas you can’t control. That’s the whole point of systems.
AI doesn’t solve this. It only exposes how weak most of us actually are at it.
What I Actually Built
I’m not saying AI is useless - far from it. I set off on this mission because the internet scared me into thinking AI would replace my EA and systems business.
My response: if AI could replace FreedUp, then I’ll use it to enhance FreedUp instead. I decided to become my own disruptor.
So I built an app called Command Center. I spent about six hours walking around my kids’ soccer fields, chatting with AI to architect the thing. It’s a Next.js app with a Supabase backend and a vector database. Here’s what it does:
On the intelligence side, it indexed every email I’ve sent or received in the last year, all my Notion items, all my Fireflies meeting transcripts, and my full calendar. It created a searchable intelligence layer - I can ask it questions about my entire professional life and it knows the answers. It even built a dynamic CRM so my EA can look up anyone I know and instantly understand the relationship.
On the execution side, I started building skills - basically AI-powered SOPs. The best one triages my inbox into four labels (action needed, follow up, interested, read later), drafts responses for action items, and creates to-dos in Notion. It even learns from my edits to improve over time.
The vision is compelling. The reality? Open Claw is incredibly slow. It breaks constantly. Its cron jobs are unreliable. You have to know how to code to keep it running. I tried to build my end-of-day shutdown routine as an AI workflow through WhatsApp, and it was painfully hard. My human assistant can do the same thing in one phone call just by reading my SOP.
The Vision That Still Makes Sense
What I’ve landed on is this: the future isn’t AI replacing your EA. It’s your EA becoming more like a chief of staff - solving complex, strategic problems - while delegating lower-level tasks to AI tools.
Just like a CEO who has both a chief of staff and an executive assistant, the human handles the relationship, the judgment, the anticipation. The AI handles the repetitive, data-heavy tasks that don’t require intuition.
My EA thinks strategically. She anticipates my needs. She manages complex client relationships. AI can’t do that. But AI can triage an inbox, prep a meeting brief, and draft a first pass at an email. Those are the tasks worth automating.
The Bottom Line
After $1,300 (if you include my personal hourly rate, more like $30k) and 150 hours, here’s what I know: treat AI like a tool, not a person. Not a helper. Not a role. A tool.
Give that tool to smart humans, train them on it, and let them be more productive, more accurate, and faster. But don’t pretend it’s going to replace the judgment, intuition, and reliability of a great human assistant.
And if you can’t delegate effectively to a human who’s intuitive, smart, and responsive - you’re definitely not going to be able to delegate to AI. The workflows have to come first. The systems have to come first. That’s always been true, and AI just makes it more obvious.
Still building. Still experimenting. I’ll keep you posted.
But if you want a fix next week - I already have it. It’s world class EAs running FreedUp’s operating system for delegation. AI is part of it, but it’s not the whole thing.
Happy Friday,
Aaron