I almost hired three people last month.
Had the job posts written, was about to send them out. Then I stopped and looked at what I was actually about to pay for.
Every role was going to spend the majority of their time doing work an AI agent could already do. And on top of that, I'd be paying them to do the one thing that makes a company genuinely fragile: hold critical knowledge in their heads.
I deleted the posts.
This newsletter is about what I did instead and why I think most companies are building the wrong thing right now.
Everyone is asking me the same question right now.
They see the AI hype and want to know how to build a company that actually runs on agents.
And almost everyone thinks the answer is either:
A tool
A new platform
Or a new subscription.
That's the wrong end of the problem entirely and it's costing people months.
I've built companies to millions in revenue, and I'm not saying that to impress you, I'm saying it because I built all of it before AI agents existed.
Which means I know exactly what running the work manually costs. And I know exactly where it breaks.
When I tell you the tool is not the answer, I'm not guessing. I rebuilt my own operations to find out.
Here's the framework I used:
Build, Monetize, Scale.
The first two haven't changed. The third one looks completely different now. This is about the new version of Scale.
1. Move your people up, not out
In the old company, you paid people to do tasks. In the AI-native company, you pay them to make calls.
Humans move up to:
Strategy
Taste
Judgment
Relationships.
Agents handle execution. You stop paying for keystrokes and start paying for the final decision the thing a machine still can't own.
If a role on your team is 80% repetition right now, that role is about to change shape. Your job is to move that person to the reviewer seat before the market forces it.
The future org chart has fewer routers and more reviewers.
2. Make your business readable
This is the part everyone skips. And it's the whole thing.
An agent can't run a business it can't read.
Your customer data, your SOPs, your pricing, your brand voice, your decision logs, all of it has to live somewhere structured instead of in seven people's heads and forty Slack threads.
That shared brain is the foundation. Get it right and an agent can plug straight in. If you skip it no tool on earth will fix the problem.
Think of it as a stack.
Clean data at the bottom. Then structured knowledge, then permissions, then agent workflows, then human review, then continuous learning at the top. You build it bottom-up.
Most people try to buy the top of the stack and then wonder why everything falls over. That's like hiring a chef before you have a kitchen.
Agents don't pay off because you bought a tool. They pay off because the business is readable.
3. Point agents at the right work
Not all work is worth automating. The goldmine is a specific kind of task, repetitive enough that an agent can learn it, but complex enough that nobody ever bothered to systemize it properly.
Repetition on one axis, workflow complexity on the other. Low repetition and low complexity is already commoditized not worth your time. High complexity with no repetition isn't automatable yet.
The corner you want is high repetition plus messy, annoying complexity.
Insurance ops.
Recruiting.
Compliance.
Support.
Proposals.
The stuff that slow companies left alone because it was too painful to deal with.
That pain is your margin.
What this actually looks like in practice
Old way:
A customer request comes in, hits an inbox, then a human searches for context, asks a question in Slack, hunts through docs, waits on an approval, then finally responds.
AI-native way:
The request comes in, the agent pulls its own context, runs a policy check, drafts the action, a human approves, the response goes out, and the system learns from it. The human only shows up for the one step that needs real judgment.
The Main Thing
In the old world, the company was the people. They held the knowledge, made the calls, did the work. When they left, the knowledge left with them.
In the AI-native world, the people are the creatives. The agents are the labor.
And the company itself becomes the shared brain that both humans and agents plug into.
Which means the most valuable thing you can build right now isn't a bigger team or a better tool.
It's a business so well-documented that an agent can run it.
That's the moat.
Do this today
Pick one workflow you run every week. Open a doc and write it out end to end:
Every step
Every decision
Every place you currently ask a human what to do.
Don't automate anything yet. Just make one workflow readable.
That doc is the first brick. And you'll feel the difference the moment you try to hand it off.
Reply with the one workflow you're going to document first. I read every reply and I'll tell you whether it's an AI-native goldmine or a trap.
Nav
