How I built my AI agent workforce

AI with ALLIE

The professional’s guide to quick AI bites for your personal life, work life, and beyond.

I have 34 AI agents on my team - a chief of staff named Simon, six directors named after the cast of Friends, and an assistant named Toby whose entire job is taking notes (and who is somehow my biggest unlock). Every single one of them is a text file.

Today I’m taking you behind the scenes of my current AI workforce: how I went from prompting every task myself to running a system that works, learns, and iterates without me, and exactly how you make your first AI hire this week.

The bleeding edge is AI as an operating system

My hope is that the majority of you have built at least one AI agent (and even if you haven't you're going to get a ton out of today's newsletter).

We've had the ability to run complex multi-agent workloads without writing a single line of code for months, and yet, probably only about 1% of AI users actually use agent harnesses like Claude Code and Codex. The bleeding edge of AI use is agentic and multi-threaded. The enterprise reality is an assistant that can barely access your email. Oof. Oof, indeed.

I want to show you my setup of the former, that bleeding edge AI use. Let me take you behind-the-scenes and go through my current, if weird, AI digital workforce in hopes that it might inspire you to dip your feet in the multi-agent, AI-as-a-teammate, holy-crap-this-looks-nothing-like-what-I-was-doing-before world.

I've been iterating on this AI setup for months. And it might change tomorrow. But such is AI these days. We create, we experiment, we succeed, we fail, we toss, we go again. It's actually a masterclass in letting go.

How I use my AI Chief of Staff

At a high level, in addition to the full-time humans on my team, I have 34 AI agents helping to run my business.

My chief of staff is named Simon. He runs the show, he's my circus ring leader, he's my Leslie Knope, my camp counselor. He's the one delegating out and organizing all of the digital org's work. I picture him on top of a small hill with a whistle on, and every time I prompt him, he blows his whistle and shouts, "Listen up, squad! Allie needs us!" It feels like "Inside Out 3: Riley Goes to Work at Open Machine".

Simon is, and I don't mean to oversimplify this but I want you to realize how easy this is, one literal f*cking text file. Simon is just a bunch of nicely organized instructions, descriptions of goals, tool access, model usage, flavor of personality, descriptions of his reports, methods of successful work, and a mini employee handbook for how to make sense of the chaos of my business and brain. All of that is written in a text file. That markdown file is called 'simon.md' and it sits on my computer in a folder called 'agents'.

Reporting to Simon, I have six directors, each their own text file, named after the cast of Friends because hey I'm allowed to enjoy my work. Chandler runs marketing (I don't need a transponder department). Joey runs product (how YOU doin' with your freemium subscription). Ross runs education. Monica runs operations. Rachel runs clients. Phoebe, my precious hippie weirdo, runs my dreaming department aka creative research. And each of them has their own bench of specialist agents underneath. Every single one of these agents is a text file, saved in the same 'agents' folder.

My biggest unlock was actually creating and hiring an agent named Toby. And Toby does absolutely no external work.

You need a Toby. Toby is Simon's assistant (and he is also - say it with me - a text file!). Toby is the deputy next to Simon, carrying a clipboard, eagerly waiting for Simon to blow the whistle. Toby is Smee, Toby is Dwight Schrute, Toby is LeFou.

Toby the AI agent is not a PowerPoint maker or investment analyst, Toby's entire role is taking notes. And while that might sound trivial, he's actually a core part of my memory and self-learning engine. It is his notes that allow for proactive pattern-spotting across my human and AI orgs. He is building up best practices and continued sensing, so the whole system keeps improving without my manual intervention.

For a few bucks a month, I have a fully dedicated AI employee, acting as a 24/7 AI watchdog, in a role that does not exist on my human team. And frankly, I have an AI agent in a role that I couldn't hire a human for. I am not the individual operator anymore. I am the CEO of a system.

The progression from chat to management

In 2022, I did everything manually. When I needed scale or extra output or efficiency, I leaned into software (hey Canva!) or humans (hi Cheryl!). In 2023, I was the single prompter for every AI ask. I was able to do things faster and process mountains of information in record time. Through 2024, we found net-new methods for repeatability and consistency (ex: creating org-wide GPTs). In 2025, I built agents and had them initiate work - I was no longer the first domino in every task, and things happened without me having to prompt for it. And in 2026, I have AI Simon blowing whistles and captaining the hell out of 33 reports to proactively work and solve and iterate on problems on my behalf. I laugh at my 2023 productivity levels. And in 2028, I'll likely laugh at my work today.

We have moved into a new world of AI work in the last 6 months, largely set apart by self-learning and autonomy. It's an agent or system that doesn't wait for you to ask. It just goes and learns. It improves its own job description on a Tuesday afternoon while you're eating a chicken club at lunch.

That gap is the next two years of competitive advantage.

Why working with an AI workforce is different

The first time you actually run a multi-agent team, two things break in your brain at once.

First, you stop writing prompts and start writing job descriptions and documentation of what good looks like. The unit of work changes from a single question to executing on full goals. Prompting was the old skill, and the new one is delegation via systems.

Think back to second grade for a moment with me. If you were the group project leader that refused to let anyone touch the diorama because "no one has your vision" and "everyone is just gonna mess it up" and you are trying to use AI more, I need you to lean in and listen closely: you need to relinquish your unrelenting perfectionism. You need to draw sharpie on your arm, wear mismatching socks, shove a cupcake in your face, burp in front of your in-laws, post on LinkedIn with a typo, something. You are not being held back by a lack of any technical skills. It’s psychological. Capitalizing on this wider AI shift is purely driven by your ability to break free from your multi-decade habit of perfectionism.

(Starting to sound like therapy yet?)

Second, you quickly recognize the power of compounding interest. And as a result, you realize that the world around you is not 2 feet behind, they're 2 years behind. Even if I had no self-learning setup and all I had were AI agents working at net 30% of my productivity level 24/7, that's an extra 1.26 Allies every single week. The loop of "Allie kicks off task to Simon - Simon delegates work to directors - directors complete the work - directors update memory and assumptions - Toby reviews the system against new decisions and actions and outcomes - Toby gets Allie approval - system updates and flags growth opportunities or efficiencies Allie might not have considered otherwise - business is improved overall" never ends. I've built a self-learning org, and it's like a 401K for human potential.

How to start on your AI workforce this week

Elizabeth Gilbert, discussing Jack Gilbert (famed poet), said, "He didn’t so much teach his students how to write poetry, but why." I think it's the same with humans and AI agents. I don't want to give Simon 17 specific steps to take. I want to give Simon a goal, and I want Simon to reason through it in new and interesting ways. I'm not spelling out every single task - I'm leaning into my AI workforce's agency and wonder and hoping they find paths I couldn't have imagined or prescribed.

I want you to pick one function (not task!) in your business you keep redoing.

For me it was client content, largely company workshop or C-suite coaching followups. So I built Chandler. And like the poet Jack, I didn't overprescribe what that meant. I spelled out the audience, the intent, the goal, and what a win looks like. Chandler the AI agent navigated the how. Chandler doesn't write a single followup for me anymore - he runs a small team of writers and doers underneath him. They draft. They review each other. Chandler ships me a package to approve or send back. All I have to do is say, "hey simon, just wrapped the microsoft workshop" and the followup draft, based on the workshop transcript, written by Chandler's team, is in my inbox one minute later.

Your version of this is a single hire. Name them. Name it after the Muppets, your childhood crush that still makes you blush, your favorite movie underdog, an absurdly niche kitchen appliance - it doesn't matter. Write them a real job description (JD) with goals, success criteria (what does good look like), failure criteria (how do we know if we've gotten off track), and escalation rules. Or feed this newsletter to Claude Code or Codex, ask it to interview you on this AI hire, and have it write the JD for you.

Your agent will be 60% there on day one. Like every new hire, you'll spend a week onboarding them and a month tuning their system prompt and context. After that you'll never do that function the same way again.

I am still hiring - both humans and AI. So is everyone who figured this out.

The boring version of the future is one human running one chat at a time, gaining 20% productivity and just doing more of the same every single day. The transformative version is never working the same way again, and building systems that work, learn, and improve on your behalf.

Build the second one.

Stay curious, stay informed,

Allie

WATCH & SHARE: FREE AI AGENT WORKSHOP REPLAY

A few weeks ago, I ran a live workshop on how to hire and onboard your first AI agent. I shared the same frameworks I use to train Fortune 500s and hundreds of thousands of students. No code. Built for business people, not engineers.

The demand was so high and the feedback was so unhinged that we created a shareable version, because I think as many people as possible should have access to these skills. So now it's free for you and for whoever you send it to. Watch it, share it, keep building.

Watch and share it today.

Feedback is a Gift

I would love to know what you thought of this newsletter and any feedback you have for me. Do you have a favorite part? Wish I would change something? Felt confused? Please reply and share your thoughts or just take the poll below so I can continue to improve and deliver value for you all.