For most of my career, the question I asked myself every morning was some version of: how do I get all of this done today?
It's the solopreneur's default setting. You are the designer, the developer, the marketer, the salesperson, the accountant, the customer support rep, the janitor. Every hat is your hat. And the hustle mythology tells you that's a badge of honor — that grinding through seventeen roles is the price of admission.
I don't ask that question anymore.
These days the question is different: what am I training my agents to do today?
I have seven of them. Seven AI agents, each responsible for a different slice of the business. One handles research and trend monitoring. One drafts content. One manages my inbox and flags what actually needs me. One does bookkeeping-adjacent work. One handles outreach. One keeps the product pipeline moving. One sits above the others and coordinates — a kind of chief of staff.
None of them are perfect. All of them are getting better.
And the strangest, most disorienting part of running this setup is how much it's changed what I actually do all day. I'm not executing work anymore. I'm shaping the team that executes the work.
When I first started building with agents, I made the mistake most people make: I thought of them as fancy automations. Plug in a prompt, get an output. One-shot tools. Glorified macros.
That framing falls apart fast the moment you try to run a real business on top of it.
An agent that just responds to prompts is a tool. A team of agents that share context, hand off work, hold memory across sessions, and improve from their own mistakes — that's an organization. And organizations behave fundamentally differently than tools.
Think of it like this: a hammer is a hammer. It doesn't learn to hit nails better. It doesn't know what the other tools in the toolbox are doing. But a carpenter on a crew? A carpenter knows the plumber is running pipe through the wall they're about to frame. A carpenter remembers that last time they used this particular wood, it split along the grain. A carpenter gets sharper every year they're on the job.
The leap in agentic workflows isn't speed. It's coordination and memory.
The mental model that unlocked this for me was treating agents less like software and more like employees. Not because they're human — they're obviously not — but because the management principles are closer to running a team than deploying a script.
A good employee needs four things to do their job well:
First, skills — they need to actually know how to do the work. For an agent, this is the prompt, the instructions, the examples, the edge cases you've walked them through.
Second, resources — they need access to the right tools. For an agent, this is APIs, webhooks, databases, document stores, the ability to read your calendar or write to your CRM. Without resources, even a skilled worker is stuck.
Third, communication — they need to know what the rest of the team is doing. An agent operating in isolation repeats mistakes the others already solved. A team of agents with shared context compounds.
Fourth, feedback — they need to know when they got it right and when they got it wrong, so they get better over time. This is where most agentic setups quietly fall apart. People build the agent, deploy the agent, and then never close the loop.
I have each of mine self-grading now. After a task, they evaluate their own output against a rubric — did they hit the brief, did they miss context, did they produce something I'd actually use — and that evaluation feeds back into how the next version of them is trained. It's not magic. It's just management basics applied to a non-human workforce.
There's a thread I keep pulling on in these posts, and it's showing up here too.
In the gatekeeper issue, I wrote about a billionaire who designed his attention around removal — filtering the world down to what actually required him. In the judgment post, I wrote about how the sharpest operators aren't the ones who know the most, they're the ones who've gotten ruthlessly good at knowing what not to do. And in the last issue, I wrote about subtraction as a design philosophy.
This is the same principle, just pointed at work itself.
The old model of solo work was additive: add more hours, add more tools, add more hustle. The agentic model is subtractive: take the execution off your plate and put it where it belongs — with a system you've trained well enough to handle it.
What's left, when the execution is gone, is the stuff that actually needs a human. Vision. Taste. Judgment. Knowing which problems are worth solving in the first place. Seeing when an agent's output is technically correct but directionally wrong. Steering the whole operation toward something that matters.
That is what I do all day now. And it's the most creative my work has ever felt.
I want to be honest about something, because I don't want this to read like a sales pitch for some agentic utopia.
My team of seven is not a team of seven senior employees. On their best days, most of them operate like a smart intern. They need direction. They miss context. They hallucinate. They do things I didn't ask for and skip things I did. I still have to review, correct, redirect — constantly.
But here's what's different from any other tool I've ever used: they get better. Not in some vague, marketing-deck sense of the word. Measurably better, week over week, because the system is designed to learn from itself. The feedback loop is the product.
An intern you train for a year and then they leave. An agent you train for a year is still there, holding every correction you ever gave them, getting sharper at the exact shape of work your business needs.
That's the part that's hard to convey to people who haven't lived inside it yet. It doesn't feel like using software. It feels like compounding.
So for me, the future of work isn't some abstract AI-will-replace-us story. It's already here, and it's weirder and more empowering than the headlines make it sound.
As a creative working alone, I can suddenly operate at the scope of a small team. Not by grinding harder. By becoming a better trainer, a better director, a better systems thinker.
The question isn't how do I do the work? anymore.
It's how do I train the team to do the work, and what am I doing with the hours that gives me back?
That second question is the one worth obsessing over. Because the agents will keep getting better whether or not I'm paying attention. What they can't do — what no amount of training will ever give them — is the vision of where the whole thing is supposed to be pointed.
That part is still mine. And honestly, I think it always will be.